Difference between revisions of "AI in 15 Minutes or Less"

From CNM Wiki
Jump to: navigation, search
(Fine tuning)
(Capability of AI)
 
(854 intermediate revisions by the same user not shown)
Line 1: Line 1:
==Introduction==
+
==What AI is about==
 
===What AI is===
 
===What AI is===
: AI stands for [[artificial intelligence]], which is the competency of a computer or a robot to perform autonomously without authoritative commands of a programmer or another system.
+
: Non-intelligent machines simply execute strict commands of their operators. Non-intelligent software follow explicit rules programmed by human developers.
  
not strictly pre-programmed tasks. Normally animals perform such tasks. They can be grouped in three [[#Task subsets of AI]]:
+
: [[#AI]] stands for [[#Artificial intelligence]], which is the competency of an artificial system to learn and perform partially or fully autonomously based on what this system such as a computer, drone, self-driving car, or robot has learned.
:# '''Obtaining data''' through listening, reading, and graphic recognition; these tasks are commonly known as [[machine learning]] (ML).
+
 
:# '''Processing data''' through its analysis and synthesis; these tasks are commonly known as [[discriminate AI]].
+
: For instance, Google-owned AlphaZero is given no knowledge beyond basic chess rules. Instead, this [[#AI model]] is prompted to play against itself and, using [[#Reinforcement learning]], to learn from its own mistakes. Win or lose, AlphaZero piles up lessons learned from its millions and millions of games. According to its creator, Demis Hassabis,
:# '''Producing results''' through representation of documents, speech, or commands, for instance, needed for autonomous driving; these tasks are commonly known as [[generative AI]].
+
:: "Actually you set off AlphaZero in the morning and it starts off playing randomly. By lunchtime, you know it's able to beat [...] most chess players. And then, by the evening, it's stronger than the world champion."
 +
 
 +
: Because of their [[#Computing power]] and [[#Scalable persistent memory]], [[#AI system]]s can learn faster and accumulate more knowledge than any human being. However, they are able to learn only in those domains and in those ways that their creators specify. No [[#Intelligent machine]] of today can learn without [[#Algorithms]].
 +
 
 +
===What AI is not===
 +
: [[#AI-powered system]]s may ''mimic'' or, more exactly, ''replicate'' human intelligence. However, they are not ''beings'', ''individuals'', ''personages'', nor ''souls''. They are tools.
 +
 
 +
: [[#Intelligent machine]]s of today lack a true understanding of its surroundings and, overall, the world at least yet. For example, [[#AI-driven system]]s can sound like humans because, similarly to ''parrots'', they learn from human speech samples.
 +
 
 +
: No single [[#AI system]] possesses consciousness, intentions, interests, motivations, subjective experiences, self-awareness or awareness of their existence. Similarly to "special police" not being regular police, as of now, artificial intelligence is not general intelligence at least yet.
  
 
===AI vs ML===
 
===AI vs ML===
Performing of "not strictly pre-programmed tasks" requires at least one specific capacity known as [[learning]]. ''Learning'' is a process of autonomous acquiring new or modifying existing data. Learning is a feature of humans and other animals. Some debate whether plants are able to learn.
+
: [[#ML]] is a separate area of studies within [[#AI]]. However, no [[#AI]] can be possible without [[#ML]]. When people say, "AI and ML", they may be emphasizing the distinction between traditional AI methods and approaches that involve machine learning techniques.
  
''AI'' is built on [[machine learning]] (ML), in which a computer or robot acquire new data autonomously, meaning without a programmer or another system's command. ''ML'' can be considered as a separate area of studies within ''AI''.
+
: Traditional [[#AI]] includes rule-based systems and expert systems, while [[#Machine learning]] relies on data-driven approaches to improve performance over time. By mentioning both ML and AI, individuals may want to acknowledge the broader context of AI and highlight the specific role that machine learning plays within that framework.
  
===What AI changes===
+
===Animals vs AI===
Artificial Intelligence (AI) is important for various reasons, and its impact extends across multiple domains. Here are some key reasons why AI is considered important:
+
: Naturally, intelligence is a capacity of animals, especially of humans. Some argue that plants demonstrate limited intelligence as well. That is why the intelligence that computers or robots may demonstrate is called "artificial".
  
    Automation of Repetitive Tasks: AI enables the automation of routine and repetitive tasks, freeing up human resources to focus on more complex and creative aspects of work. This can lead to increased efficiency and productivity in various industries.
+
: Parents teach their kids, while giving them initial instructions and exposing them to the real world to learn. To train an [[#AI system]], their creators supply this system with [[#Data]] and [[#Algorithms]] to process it. No parents can predict everything their kids do; similarly, designers of [[#AI model]]s cannot anticipate everything their creations would produce.
  
    Enhanced Decision Making: AI systems can process and analyze vast amounts of data at speeds that are beyond human capabilities. This enables better decision-making by providing insights, patterns, and predictions based on data analysis.
+
: Nevertheless, there is a distinctive difference between raising children and training of machines. Parental controls over what their kids learn and over patterns of their learning are limited. Developers of [[#AI model]]s today have full control over [[#Algorithms]] that the machines utilize to learn and perform, as well as over [[#Data]] provided during pre-training.
  
    Improved Efficiency and Productivity: AI technologies, such as machine learning algorithms, can optimize processes, streamline workflows, and improve overall efficiency. Businesses and industries can benefit from AI-driven solutions to enhance productivity and reduce costs.
+
==Capability of AI==
 +
===What AI can===
 +
: By definition, [[#AI]] can perform at least one task autonomously, where "autonomously" refers to the limited ability of an [[#AI system]] to function independently without external control or direct human intervention.
  
    Innovation and Creativity: AI has the potential to contribute to innovation by providing new ways of solving problems and generating novel ideas. For example, AI-powered algorithms can assist in creative tasks such as design, music composition, and content creation.
+
: [[#AI system]]s have the capability to execute tasks that are not exclusively programmed in advance. They can be grouped in three task sets:
 +
:# '''Partially-autonomous acquisition''' of data through listening, reading, and graphic recognition. These tasks are commonly known as [[#Machine learning]] ([[#ML]]).
 +
:# '''Partially-autonomous discrimination''' of data through its analysis and classification. The word "discrimination" originates from the Latin verb "discriminare," which means "to distinguish" or "to separate." Those [[#AI model]]s that perform these tasks are commonly known as [[#Discriminative model]]s.
 +
:# '''Partially-autonomous generation''' of results through delivery of produced outputs such as documents and commands, for instance, needed for autonomous driving. Those [[#AI model]]s that perform these tasks are commonly known as [[#Generative model]]s.
  
    Personalization and Customization: AI systems can analyze user behavior and preferences to offer personalized experiences in areas like online shopping, content recommendations, and targeted marketing. This enhances user satisfaction and engagement.
+
: Generally speaking, autonomy implies some level of self-governance and the ability to adapt to changing circumstances without relying on continuous commands, guidance, or instructions from external sources.
  
    Healthcare Advancements: AI is making significant contributions to the healthcare industry, from medical image analysis to drug discovery. AI technologies can assist in diagnosing diseases, predicting patient outcomes, and personalizing treatment plans.
+
===What AI can't yet===
 +
: As of today, [[#AI system]]s are only partially autonomous, lacking the ability to act beyond their specified functions and requiring human guidance for contextual understanding, goal-setting, and supervision. Currently, [[#AI]]:
 +
:* '''Can't perform without [[#Algorithms]]''' and, often, training: [[#AI model]]s encounter challenges when adapting to unseen contexts due to their reliance on the training data, struggling with entirely new or unforeseen situations that were not part of their training set and lacking the ability to generalize knowledge like humans. While AI can generate novel outputs based on learned patterns, it falls short of true creative thinking, imagination, and inspiration, as it cannot form unique concepts beyond the confines of its training data.
 +
:* '''Have no common or conventional senses''': AI lacks inherent ethics. Ethical concerns arise from human developers' decisions and biases in training data, unintentionally perpetuating biases. Despite potency in tasks, AI faces challenges in common sense reasoning, interpreting implicit meanings, grasping context, and making inferences beyond training data.
 +
:* '''Have no emotion, intuition, or feeling''': AI lacks the capacity for intuition or gut feelings, as human intuition often involves subconscious processing and experiences that are difficult to reproduce in machines. While certain AI applications may simulate emotional responses or recognize human emotions, such simulations are rooted in pattern recognition and do not constitute a true understanding of emotions.
 +
: As of today, [[#AI system]]s operate within the parameters set by their programming with [[#Algorithms]] and training.
  
    Improved Customer Service: AI-powered chatbots and virtual assistants are increasingly used in customer service to provide quick and accurate responses to queries. This improves customer satisfaction and can handle a large volume of inquiries simultaneously.
+
===AI vs AGI===
 +
: As of now, [[#AI]] refers to limited intellectual capacity called [[#Narrow AI]]. [[#Artificial intelligence]] of today can perform partially autonomously narrow, specific tasks such as chatbots, facial and speech recognition, recommendation systems, self-driving cars, and virtual assistants.
  
    Security and Fraud Detection: AI is instrumental in enhancing cybersecurity by detecting and preventing cyber threats in real-time. It is also used for fraud detection in financial transactions and other domains.
+
: On the contrary, [[#Artificial general intelligence]] or [[#AGI]], if developed, would be able to perform any intellectual task that a human can do. [[#General AI]] is in the research phase and is not commercially available.  
  
    Autonomous Systems: AI plays a crucial role in the development of autonomous systems, including self-driving cars, drones, and robotics. These systems can perform tasks without human intervention, contributing to safety and efficiency.
+
: In 2019, Elon Musk expressed his belief that [[#AGI]] will be possible by 2029. In the same year, China-based AI visionary Kai-Fu Lee said in an interview that [[#AGI]] would be developed, "not within the next 30 years and possibly never".  
  
    Scientific Research: AI aids scientific research by analyzing complex data sets, simulating experiments, and identifying patterns in large datasets. This accelerates the pace of scientific discovery.
+
: If researchers are able to develop an AGI system, it could have significant implications for society, including creation of machines that are more intelligent than humans, which could pose ethical and safety concerns.
  
While AI offers numerous benefits, it is essential to consider ethical and societal implications, including concerns related to bias, privacy, and job displacement. Responsible development and deployment of AI technologies are crucial to realizing the full potential of AI while addressing associated challenges.
+
===Growth of AI capacity===
  
Artificial Intelligence (AI) is transforming the world in various ways, influencing diverse sectors and aspects of human life. Here are several ways in which AI is changing the world:
+
==Impact on jobs==
 +
[[How Safe Is My Job]]
  
    Automation of Tasks: AI has enabled the automation of routine and repetitive tasks across industries, leading to increased efficiency and productivity. This ranges from manufacturing processes to administrative tasks, freeing up human workers to focus on more complex and creative aspects of their jobs.
+
==AI system==
 +
An [[#AI system]] refers to a [[#System]] such as a computer, drone, self-driving car, or robot that can perform one or more tasks autonomously, without explicit human command. Some people refer [[#AI system]]s to software only and use terms such as [[#AI-driven system]]s, [[#AI-powered system]]s, or [[#Intelligent machine]]s to indicate not-only-software systems. However, those terms are often used interchangeably.
  
    Data Analysis and Insights: AI excels at processing and analyzing vast amounts of data quickly. This capability is harnessed in various fields, including business, finance, healthcare, and research, to derive insights, identify patterns, and make data-driven decisions.
+
Literally, [[#AI]] is software. Any software requires hardware, which stands for physical devices that allows for software to run. Any, robotic or non-robotic, [[#AI system]] needs (a) [[#Computing power]] to compute, (b) [[#Persistent memory]] to store [[#Data]], (c) actuators, dashboards, keyboards, sensors, or other [[#AI peripheral]]s, as well as (d) connectors and wires to connect the system's components.
  
    Enhanced Decision-Making: AI systems, particularly those based on machine learning, can analyze data to provide valuable insights for decision-makers. This can lead to more informed and effective decision-making in areas such as business strategy, finance, and public policy.
+
Besides [[#AI model]]s and hardware, [[#AI]] needs [[#AI agent]]s, [[#Algorithms]], [[#Data]], and the software to enable those. So, when we say [[#AI]], we often refer to some combination of software and hardware pieces. Depending on the type and complexity of the task, [[#AI system]]s can use different combinations of hardware and software components.
  
    Personalization and User Experience: AI technologies power personalization in products and services, tailoring experiences based on individual preferences and behaviors. This is evident in personalized recommendations on streaming platforms, targeted marketing, and customized user interfaces.
+
===AI agent===
 +
: An AI agent is any software or hardware component of an [[#AI system]] that is responsible for:
 +
:# '''Input data intake''' through perceiving the system's environment. To collect data, AI agents often have [[#AI peripheral]]s.
 +
:# '''System's decision''' through using algorithms, rules, or machine learning models to analyze the information they receive and make decisions based on that analysis. The decision-making process is typically aimed at achieving a specific goal or solving a particular problem.
 +
:# '''Output delivery''' such as taking actions to achieve specific goals. Once an [[#AI system]] made a decision, the AI agent takes actions in the environment. This could involve physical actions in the case of a robotic agent or virtual actions in the case of software-based agents.
  
    Healthcare Advancements: AI is revolutionizing healthcare through applications like medical imaging analysis, predictive analytics, drug discovery, and personalized medicine. AI technologies contribute to more accurate diagnoses, treatment planning, and improved patient outcomes.
+
: Based on their level of autonomy and learning capability, AI agents may be divided in (a) [[reactive agent]]s that follow predefined rules and respond to specific situations without learning from experience and (b) [[intelligent agent]]s that are capable to learn from experience, adapt their behavior based on feedback, improve performance over time, and/or acquire new knowledge to handle different situations. Intelligent agents are [[#AI system]]s on their own.
  
    Autonomous Systems: AI plays a crucial role in the development of autonomous systems, including self-driving cars, drones, and robots. These systems have the potential to transform transportation, logistics, and various industries by reducing the need for human intervention.
+
===Algorithms===
 +
: In the context of [[#AI system]]s, [[#Algorithm]]s are the rules or instructions that [[#AI system]]s follows to complete their tasks. In the broader sense, an algorithm is a step-by-step set of directives, procedures, and/or regulations that dictate how a specific task or computation should be performed.
  
    Natural Language Processing and Communication: AI-powered natural language processing (NLP) enables machines to understand and generate human language. This is evident in virtual assistants, chatbots, and language translation services, improving communication and accessibility globally.
+
: [[#AI system]]s normally utilize several [[#Algorithms]]. Generally speaking, the more complex [[#AI system]] is, the more [[#Algorithms]] it tends to utilize. The [[#Algorithms]] can be divided in two categories:
 +
:# Those that are used in [[#Model development]].
 +
:# Those that are used in [[#Inference]] or making predictions on new, unseen data. The [[#Inference]] algorithms, particularly, are used to process data and perform tasks that [[#End-user]]s prompt.
  
    Scientific Discovery: AI accelerates scientific research by analyzing complex datasets, simulating experiments, and identifying patterns in scientific data. This contributes to advancements in fields such as physics, chemistry, and biology.
+
: [[#Algorithms]] can be based on logic, optimization, or statistical techniques. Data scientists and engineers who specialize in [[#AI]] and [[#ML]] design mathematical models, develop learning algorithms, implement them using programming languages, test their predictions, and adjust them based on the testing.
  
    Cybersecurity and Fraud Detection: AI is employed in cybersecurity to detect and prevent cyber threats in real-time. It also aids in identifying patterns indicative of fraudulent activities in financial transactions and other domains.
+
===Computing power===
 +
: In the context of [[#AI system]]s, [[#Computing power]] refers to the capabilities of the hardware processors that perform the tasks and computations associated with artificial intelligence (AI) applications.
 +
:* [[#Central processing unit]]s ([[#CPU]]s) are normally used in simple models on small datasets, inference on low-resource devices, non-deep learning algorithms, and demonstrations. Because of their limited number of cores, they are not sufficient in running more complex computations in [[#AI]].
 +
:* [[#Graphics processing unit]]s ([[#GPU]]s) offer significant advantages in terms of accelerating various AI processes in comparison with [[#CPU]]s. [[#GPU]]s have hundreds or even thousands of cores, which allow them processing their tasks simultaneously in parallel. Although [[#GPU]]s were initially designed for graphics processing, they make early [[#Deep learning]] possible. Large-scale models with millions or billions of parameters require more powerful hardware accelerators than [[#CPU]]s.
 +
:* [[#Tensor Processing Unit]]s ([[#TPU]]s) are the newest category of processors in [[#AI]]. [[Google]] developed them specifically for [[#Deep learning]]. However, because of [[#TPU]]s' significant costs, they have been only used for some of the most recent and advanced models, such as GPT-4 and CodeLlama.
 +
: As of early 2024, the statistics is not available, but it seems that [[#GPU]]s are still the most widely used hardware processors for AI computing, followed by [[#TPU]]s, which are gaining popularity and adoption. [[#CPU]]s are likely to have a very low percentage of AI computing, as they are not efficient or powerful enough for most AI tasks.
  
    Education and Skill Development: AI technologies are being used in education for personalized learning experiences, intelligent tutoring systems, and skill development platforms. This can cater to individual learning styles and pace.
+
: In the meanwhile, once trained, any [[#AI model]] needs a [[#Deployment infrastructure]] to function in real-world applications. Efficient hardware and software [[#Training infrastructure]] is necessary for training large-scale AI models. Without the appropriate infrastructure, training times can be prohibitively long, hindering the development process.
  
    Environmental Monitoring and Conservation: AI is utilized in environmental monitoring, climate modeling, and conservation efforts. It helps analyze large datasets related to climate change, biodiversity, and natural resource management.
+
===Data===
 +
: In the context of [[#AI system]]s, [[#Data]] is the raw material that [[#AI system]]s use to learn from and make predictions. [[#Data]] can be [[#Structured data]] (such as categories, labels, or numbers) or [[#Unstructured data]] (such as audio, images, or text). [[#End-user]]'s input such as a [[#Prompt]] is a part of [[#Data]] as well.
  
While AI brings about significant positive changes, it also raises challenges such as ethical considerations, job displacement, bias in algorithms, and privacy concerns. It is essential to address these challenges responsibly to ensure that AI benefits society as a whole.
+
[[#Persistent memory]]
  
===What AI is not===
+
===Scalable persistent memory===
 +
: Human memory is not a persistent storage system and is subject to decay and changes over time.
  
==AI task subsets==
+
Data intake and storage
 +
: Including code, image, text, video, and voice acquisition. AI systems can accept large amounts of data faster and more accurately than humans. For example, AI can identify toys, faces, animals, and even diseases from images. They can also understand and transcribe human speech with high accuracy.
  
Artificial Intelligence (AI) is a broad and multidisciplinary field that encompasses various subfields and branches. Here are some key branches of AI:
+
==AI model==
 +
An [[#AI model]] is the knowledge abstraction or representation that makes an [[#AI system]] intelligent.
  
    Machine Learning (ML): ML is a subset of AI that focuses on the development of algorithms and statistical models that enable computers to perform tasks without explicit programming. It includes supervised learning, unsupervised learning, and reinforcement learning.
+
own knowledge abstractions called [[#AI model]]s.
  
    Natural Language Processing (NLP): NLP involves the interaction between computers and human (natural) languages. It includes tasks such as speech recognition, language translation, and sentiment analysis.
 
  
    Computer Vision: This branch enables machines to interpret and make decisions based on visual data. Computer vision is used in image and video recognition, object detection, facial recognition, and more.
+
[[#Model development]]
  
    Robotics: AI in robotics involves creating intelligent machines capable of performing tasks in the physical world. It combines elements of computer science, engineering, and mechanical design.
+
===Model architecture===
 +
: [[#AI system]]s often deploy multiple models and their combination is called [[#Model architecture]].
  
    Expert Systems: Expert systems are computer programs designed to mimic the decision-making abilities of a human expert in a specific domain. They use knowledge bases and inference engines to make decisions.
+
: Different models make predictions, classifications, or decisions based on new or unseen data. Some models work with input agents, some do with decision agents, and same work with output agents. For instance, in [[#GPT]],
 +
:* A dedicated tokenization model chunks [[#End-user]]'s request in smaller tokens.
 +
:* [[#Discriminative model]]s process the chunked tokens. One model maps it to a high-dimensional vector, known as a token embedding. Another model adds to this embedding token's position in a prompt, etc.
 +
:* Finally, [[#Generative model]]s are used to generate the output. For instance, one model may suggest the output format, another may so its content, the third may make sure that the output fits the audience, etc.
  
    Knowledge Representation and Reasoning: This branch focuses on how to represent information about the world in a form that a computer system can utilize to solve complex tasks. It also deals with the reasoning processes that derive conclusions from that information.
+
: Any [[#AI model]] generalizes its learning from the training data to perform tasks on new, similar data. That way, the models can capture patterns, relationships, or features that are relevant for a specific task or domain.
  
    Planning and Decision Making: This involves creating algorithms and systems that can make decisions and plan actions in dynamic environments. It's crucial for AI systems to make intelligent choices based on available information.
+
===Large language model===
  
    Game Playing: AI has been applied to playing games, ranging from traditional board games like chess and Go to modern video games. This involves creating algorithms that can strategize and compete against human or other AI opponents.
+
==Workflow of AI system==
 +
: For example, when a user prompts a [[#Generative Pre-trained Transformer]] or [[#GPT]] to answer his or her question, this AI system:
 +
:# '''Transforms''' the unstructured human prompt, while encoding it into a set of structured machine requests. The machine requests identify the prompted answer format, the topic or topics to address, etc. To create structured requests, the system compares the human request with the prompt dataset.
 +
:# '''Studies''' those areas of its large dataset that are relevant to the request. To create that dataset, the system receives the data, usually, an enormous volume of data, in advance. This process is known as ''pre-training''.
 +
:# '''Generates''' the response, while selecting most likely words in a most likely sequence from large dataset.  
  
    Machine Perception: Machine perception is about endowing machines with the ability to interpret and understand the world through senses like vision and hearing. It includes tasks such as object recognition, speech recognition, and gesture recognition.
+
==Model development==
 +
[[#Algorithms]], both learning and optimization, are integral to creating models in machine learning. Learning algorithms such as ones for [[#Training]], [[#Evaluation]], and/or [[#Hyperparameter tuning]] enable models to discover patterns in data. [[#Optimization]] algorithms adjust model parameters for optimal performance during training.
  
    Swarm Intelligence: Inspired by the collective behavior of social insects, swarm intelligence involves the study of systems where a large number of simple agents work together to solve complex problems.
+
===Training===
 +
: During the training phase, an algorithm is employed to optimize the parameters of a model. This involves presenting the model with a dataset, allowing it to learn patterns and relationships within the data. The algorithm adjusts the model's parameters iteratively to minimize the difference between its predictions and the actual outcomes in the training data.
  
    Evolutionary Computation: This branch involves algorithms inspired by the process of natural selection. Genetic algorithms, for example, are optimization algorithms that use principles of natural selection to evolve solutions to problems.
+
===Optimization===
 +
: Optimization algorithms are often used to fine-tune the parameters of a model to improve its performance. Gradient descent is a common optimization algorithm that adjusts the model's parameters based on the gradient of the loss function with respect to those parameters.
  
These branches often overlap, and advancements in one area can benefit others. AI is a rapidly evolving field, and new branches and interdisciplinary approaches continue to emerge.
+
===Evaluation===
 +
: Algorithms are also used to evaluate the performance of models. Metrics such as accuracy, precision, recall, and F1 score are calculated using specific algorithms to assess how well the model generalizes to new data.
  
==Applications of AI==
+
===Hyperparameter tuning===
 +
: Algorithms play a role in the process of hyperparameter tuning, where the values of hyperparameters (settings not learned during training) are adjusted to find the optimal configuration for the model.
  
===Narrow===
+
==AI peripheral==
: [[Narrow AI]] is designed for specific tasks,  
+
In the context of [[#AI system]]s, peripheral devices or [[#AI peripheral]]s enable the system's interaction with the physical world. The peripherals represent the system's hardware beyond their core computing. To interact with the digital world, [[#AI system]]s utilize [[#API]]s.
  
Narrow AI, also known as Weak AI, refers to artificial intelligence systems that are designed and trained for a specific task or a narrow set of tasks. Unlike general artificial intelligence, which would possess the ability to understand, learn, and perform any intellectual task that a human being can, narrow AI is limited to a predefined function.
+
The peripheral devices interface with or connect to [[#AI agent]]s to facilitate their communication with their their physical environment. [[End-user]]s of [[#AI system]]s use the peripherals to work with these systems.
  
Narrow AI systems excel at performing well-defined and specific tasks, such as image recognition, language translation, speech recognition, and playing board games like chess or Go. These systems are not capable of generalizing their knowledge or skills to perform tasks outside their designated scope.
+
===Input devices===
 +
: AI input devices enable input from the physical world into [[#AI system]]s.
 +
:* '''Manipulators''' such as buttons, joysticks, keyboards, mouses, steering wheels, etc. enable human manual input into [[#AI system]]s. Utilizing those, [[#End-user]]s may manually input commands, queries, or parameters.
 +
:* '''Sensors''' such as accelerometers, cameras, microphones, radars, scanners, and etc. enable automatic input into [[#AI system]]s.
  
In contrast to narrow AI, general AI would have the capability to understand and adapt to a wide range of tasks, similar to the cognitive flexibility of a human being. As of now, we have not achieved general AI, and most AI applications in use today are examples of narrow or specialized AI.
+
===Output devices===
 +
: AI output devices enable output from [[#AI system]]s into the physical world. To impact the real world, a few devices can be used such as:
 +
:* '''Actuators''' prompt physical actions of motors, robotic arms, or other mechanisms when [[#AI system]]s make such decisions.
 +
:* '''Dashboards and displays''' (except [[#Two-way portals]]) present or display relevant information or insights. Dashboards visually represent data and information in a user-friendly way that allows users to monitor and analyze key metrics or performance indicators. They provide a centralized and easily digestible view of information. Dashboards and displays are not inherently designed for two-way communication, direct interaction, or data input from users. Touch-screen displays and similar devices belong to [[#Two-way portals]].
 +
:* '''Printers''':  Output devices that produce tangible results based on the AI system's outputs. This could include generating reports, images, or other forms of information.
  
===General===
+
===Two-way portals===
general AI, which can learn and perform any task that humans can. AI is a rapidly evolving field that has many applications and challenges in the modern world.
+
: Two-way portals are interactive platforms where [[#End-user]]s can both consume and contribute content. While utilizing interactive websites, collaboration platforms, and customer portals, the users can not only view information, but also actively participate by submitting data, making updates, or engaging in communication.
  
General AI, or Artificial General Intelligence (AGI), refers to a type of artificial intelligence that possesses the ability to understand, learn, and apply knowledge across a broad range of tasks at a level comparable to human intelligence. Unlike narrow AI, which is designed for specific tasks, general AI would have the capacity for cognitive functions similar to those of a human being.
+
==Key concepts of AI==
 +
AI concepts overlap; advancements in one area can benefit others. AI is a rapidly evolving field, and new concepts and interdisciplinary approaches continue to emerge.
  
Key characteristics of General AI include:
+
===Computer vision===
 +
: [[#Computer vision]] involves the use of algorithms and artificial intelligence to enable computers to interpret and understand visual information from the world, typically through images or videos. It can be employed for various tasks, including object recognition, image classification, facial recognition, and even autonomous navigation for vehicles.
  
    Cognitive Flexibility: The ability to understand and adapt to various types of tasks without the need for explicit programming.
+
===Deep learning===
 +
: [[#Deep learning]] is a subset of machine learning that involves neural networks with multiple layers (deep neural networks) to automatically learn hierarchical representations of data. It excels in tasks such as image and speech recognition, natural language processing, and complex pattern recognition, enabling systems to automatically extract intricate features and make high-level abstractions from raw input data.
  
    Learning and Reasoning: General AI would be capable of learning from experience, reasoning through problems, and making decisions based on acquired knowledge.
+
===Expert systems===
 +
: [[#Expert system]]s emulate human expertise by using a knowledge base and a set of rules to make decisions or solve problems within a specific domain. They can be applied in areas such as diagnosis, decision-making, and troubleshooting, providing intelligent solutions by mimicking the decision-making process of human experts.
  
    Problem Solving: The ability to tackle complex problems and find solutions in a manner similar to human problem-solving.
+
===Fine-tuning===
 +
: [[#Fine-tuning]] in machine learning involves adjusting pre-trained models to better suit specific tasks or datasets, leveraging existing knowledge for improved performance in specialized applications such as image classification or natural language processing. It allows for optimization without starting the training process from scratch, making it a valuable technique for enhancing model efficiency and effectiveness.
  
    Adaptability: General AI would be able to apply knowledge gained in one domain to another, showcasing adaptability and versatility.
+
===Machine learning===
 +
: [[#Machine learning]] is a subset of artificial intelligence that involves the use of algorithms to enable computers to learn patterns and make predictions or decisions from data without explicit programming. It can be applied to various tasks, including image recognition, recommendation systems, and predictive analytics, allowing systems to improve their performance over time through experience with new information.
  
Achieving General AI is a significant and complex challenge, as it requires the development of algorithms and systems that can mimic the broad spectrum of cognitive abilities found in humans. As of my last knowledge update in January 2022, we have not yet achieved true General AI, and most AI systems in use are specialized or narrow AI designed for specific applications. Researchers and scientists continue to explore ways to advance AI capabilities and work toward the goal of achieving General AI in the future.
+
Machine learning is a specific approach within the field of AI that involves creating algorithms and models that enable machines to learn patterns and make decisions without being explicitly programmed.
  
==AI Concepts==
+
===Natural language processing===
key AI concepts
+
: [[#Natural language processing]] ([[#NLP]]) involves the use of computational algorithms to analyze and understand human language, allowing computers to interpret, generate, and respond to text or speech. NLP can be applied to tasks such as sentiment analysis, language translation, and chatbot development, enabling machines to interact with and comprehend human communication.
  
===Machine learning===
 
Explain how it works and what it can do.
 
===Deep learning===
 
Explain how it works and what it can do.
 
 
===Neural networks===
 
===Neural networks===
Explain how it works and what it can do.
+
: [[#Neural networks]] are computational models inspired by the human brain's structure, composed of interconnected nodes organized into layers. They can learn complex patterns from data, making them versatile for tasks such as image and speech recognition, natural language processing, and solving intricate problems across various domains.
===Natural language processing===
+
 
Explain how it works and what it can do.
+
===Reinforcement learning===
===Computer vision===
+
: [[#Reinforcement learning]] is a machine learning paradigm where an agent learns to make decisions by interacting with an environment and receiving feedback in the form of rewards or penalties. It can be applied to tasks such as game playing, robotic control, and optimization problems, enabling systems to learn optimal strategies through trial and error.
Explain how it works and what it can do.
+
 
 +
===Robotics===
 +
: [[#Robotics]] involves the design, creation, and operation of robots, which are autonomous or semi-autonomous machines capable of performing tasks in the physical world. Robotics can be applied in various fields, including manufacturing, healthcare, exploration, and logistics, to automate processes, assist humans, and execute complex operations in environments that may be hazardous or impractical for humans.
 +
 
 +
===Supervised learning===
 +
: [[#Supervised learning]] is a machine learning paradigm where the algorithm is trained on a labeled dataset, meaning it learns from input-output pairs to make predictions or classifications on new, unseen data. It can be applied to tasks such as spam detection, image recognition, and speech recognition, where the algorithm generalizes patterns from known examples to make accurate predictions on new instances.
 +
 
 +
==Discriminative AI==
 +
[[#Discriminative AI]] refers to [[#Discriminative model]]s. They are designed to (a) classify input data and/or (b) predict an outcome based on input data.
 +
 
 +
===Acknowledged speech===
 +
: [[#Discriminative model]]s can acknowledge spoken language and convert it into text. Speech recognition systems learn to map audio features to words or phrases, making them useful in applications like virtual assistants and transcription services.
 +
 
 +
===Classified code or text===
 +
: [[#Discriminative model]]s can categorize topics, classify scripts, both code and text, detect spam, and analyze sentiments in [[#Natural language processing]] ([[#NLP]]).
 +
 
 +
===Detected activity===
 +
: [[#Discriminative model]]s can detect activities in cybersecurity, finance, and retail by analyzing patterns, anomalies, and trends in data to identify and categorize different cyber threats, financial transactions, and consumer behaviors. To detect fraud, these models learn patterns of normal behavior and can identify deviations that may indicate fraudulent transactions.
 +
 
 +
===Diagnosed condition===
 +
: [[#Discriminative model]]s can classify symptoms, diagnose medical conditions, and predict patient outcomes based on medical images and other patient's data. [[#Convolutional neural network]]s ([[#CNN]]s) are often employed for classification tasks performed on [[#Recognized image]]s in healthcare, helping radiologists and clinicians with diagnostics.
 +
 
 +
===Recognized image===
 +
: [[#Image recognition]], a subset of [[#Computer vision]], involves using artificial intelligence to interpret and understand the content of images. Besides recognizing faces, gestures, and illustrations, image recognition can be used to:
 +
:* '''Analyze''' satellite image in agriculture for crop monitoring, urban planning for infrastructure development, and environmental monitoring for assessing changes in ecosystems.
 +
:* '''Convert''' handwritten characters or words into digital formats.
 +
:* '''Detect''' unusual activities in finance and security, irregular changes, unauthorized objects in secure areas, tumors in medical images such as MRI, CT scans, etc..
 +
:* '''Identify''' defects, anomalies, or inconsistencies in products either in manufacturing or maintenance via automated optical inspection (AOI).
 +
:* '''Scan''' barcodes and QR codes. This is commonly used in retail for product identification, ticketing systems, and inventory management.
 +
:* '''Tag''' people automatically in social media photos.
 +
:* '''Track''' customer movements and behaviors within a store, providing insights into customer preferences and helping retailers optimize store layouts and product placements.
 +
 
 +
: Image recognition often include [[#Object detection]] as a specific task within its scope; while image recognition generally refers to identifying and categorizing objects or patterns in images, object detection specifically involves locating and delineating individual objects within an image. Obviously, before an object can be recognized, it should be detected.
 +
 
 +
==Generative AI==
 +
[[#Generative AI]] refers to [[#Generative model]]s that can create new content, script, etc. Two primary approaches are the following.
 +
 
 +
===New object synthesis===
 +
 
 +
: AI-powered new object synthesis involves generating new content, text, code, image, etc. from scratch in order to create [[#Synthesized objects]].
 +
 
 +
===Feature transfer===
 +
: AI-powered feature transfer involves application of existing features of one objects to different objects in the context of generative AI in order to apply [[#Transferred features]].
 +
 
 +
==Transferred features==
 +
AI-powered feature transfer refers to the process of applying the behaviors, style, nonverbal communications, and/or other features of one object to another while preserving the content of the second object. In feature transfer, the model extracts specified features from one object and applies them to another.
 +
 
 +
This process can also be called "neural feature transfer." The terms such as "style transfer", "nonverbal cues transfer", etc. can describe specific processes.
 +
 
 +
===Deepfake===
 +
: [[#Deepfake]]s are realistic forgeries combining [[#Generated image]], [[#Generated video]], and/or [[#Synthesized speech]]. StyleGAN (Generative Adversarial Network) is one of the leading companies that specialize in those applications.
 +
 
 +
===Transferred artistic style===
 +
: [[#Neural style transfer]] involves using a pre-trained convolutional neural network to blend the content of one image with the style of another. Challenges include computational costs, potential artifacts in generated images, and the need for careful parameter tuning to achieve visually pleasing results.
  
==AI Implementations==
+
===Transferred nonverbal cues===
Show some demonstrations of AI implementations, such as generative AI, chatbots, image recognition, speech synthesis, etc. Explain how they are built and what challenges they face.
+
: [[#Deep Video Portraits]] is a technique that enables photo-realistic re-animation of portrait videos using only an input video. It manipulates or synthesizes video content with a focus on accurately transferring facial expressions and movements from one person to another in a video. It has applications in video editing, movie post-production, visual effects, visual dubbing, virtual reality, and telepresence.
===Generative creations===
 
Generative AI refers to AI systems that can create new content, such as images, text, music, or even entire pieces of art. Here are a few examples of generative AI implementations:
 
  
1. **Text Generation: OpenAI's GPT Models**
+
==Synthesized objects==
  - OpenAI's GPT-3 (Generative Pre-trained Transformer 3) is an example of a powerful generative language model. It can generate coherent and contextually relevant text based on a given prompt. Developers can use GPT-3 to create chatbots, content generation tools, and more. For instance, it can generate creative writing, answer questions, or even write code snippets.
+
AI-powered synthesis involves creating entirely new code, image, text, sound, video, or even entire pieces of art from scratch. In this case, "scratch" refers to the [[#AI model]]s that are created as a result of [[#Training]].
  
2. **Image Generation: DeepDream and DALL-E**
+
Challenges in AI generation encompass avoiding inaccuracies, biases, and ethical concerns, whether in text, images, videos, music, or speech. Key hurdles include ensuring realism, coherence, and addressing potential biases in training data for responsible and inclusive generation across diverse domains.
  - Google's DeepDream is an example of generative AI for images. It enhances and modifies images in a dreamlike manner. Another example is DALL-E, also from OpenAI, which is designed to generate images from textual descriptions. For instance, you could ask it to generate images of "a two-story pink house shaped like a shoe."
 
  
3. **Style Transfer: Neural Style Transfer**
+
===Generated code or text===
  - Neural Style Transfer is a technique that uses generative AI to apply the artistic style of one image to another. This can be applied to photos, videos, or other visual content. The algorithm separates and recombines the content and style of images to create visually striking results.
+
: AI script generators are constructed using deep learning architectures like [[#Recurrent neural network]]s ([[#RNN]]s) or transformers, trained on diverse code and/or textual datasets. As of early 2024, the leading script generators are OpenAI's GPT Models (ChatGPT-3 and ChatGPT-4), Google's Bard, and HummingFaces.
  
4. **Music Generation: OpenAI's MuseNet**
+
===Generated image===
  - OpenAI's MuseNet is an AI model designed for generating music. It can compose music in various styles and genres, combining elements from different musical traditions. This allows for the creation of unique and original musical compositions.
+
: AI image generators are typically constructed using deep neural networks, such as [[#Generative adversarial network]]s ([[#GAN]]s) or [[#Variational autoencoder]]s ([[#VAE]]s), trained on extensive datasets of images. Some image generators are designed to generate images from textual descriptions. For instance, you could ask DALL-E to generate images of "a one-story blue house shaped like a flower." As of early 2024, the leading image generators are OpenAI's DALL-E, Google's DeepDream, and Midjourney.
  
5. **Face Generation: StyleGAN and Deepfakes**
+
===Generated video===
  - StyleGAN (Generative Adversarial Network) is a model that excels at generating realistic human faces. It has been used in applications like deepfakes, where AI is employed to create realistic videos of individuals saying or doing things they never did. While deepfakes raise ethical concerns, they also demonstrate the capabilities of generative AI in manipulating visual content.
+
: AI video generators are built using deep learning techniques, particularly through the utilization of [[#Generative model]]s such as [[#Generative adversarial network]]s ([[#GAN]]s) or [[#Variational autoencoder]]s ([[#VAE]]s) trained on large datasets of videos.
  
6. **Video Generation: Deep Video Portraits**
+
===Synthesized music===
  - Deep Video Portraits is a technology that uses generative AI to transfer the facial expressions and movements of one person onto another in a video. This can be used for applications such as video conferencing or virtual avatars.
+
: AI music generators are built using machine learning algorithms, often employing techniques such as [[Recurrent neural network]]s ([[#RNN]]s) or transformer models trained on vast datasets of musical compositions. As of early 2024, the leading music generator is OpenAI's MuseNet.
  
7. **Interactive Art: AI Dungeon**
+
===Synthesized speech===
  - AI Dungeon is an example of an interactive storytelling application. It uses a generative language model to respond dynamically to user inputs, creating a unique and evolving narrative based on user choices. It demonstrates how generative AI can be applied to create interactive and personalized experiences.
+
: [[Speech synthesis]], also known as [[#Text-to-speech]] ([[#TTS]]), is a technology that converts written text into spoken words. AI speech generators are built using deep learning techniques, particularly with [[#Recurrent neural network]]s ([[#RNN]]s) or [[#Transformer model]]s, trained on large datasets of human speech.
  
These examples showcase the diversity of generative AI applications across different domains. It's important to note that while generative AI has the potential for creativity and innovation, it also poses challenges related to ethics, bias, and the responsible use of technology.
+
==Assistants and chatbots==
  
===Chatbots===
 
 
Chatbots are applications that use natural language processing (NLP) and machine learning to engage in conversations with users. They find applications in customer support, information retrieval, and various other domains. Here are a few examples of chatbot implementations:
 
Chatbots are applications that use natural language processing (NLP) and machine learning to engage in conversations with users. They find applications in customer support, information retrieval, and various other domains. Here are a few examples of chatbot implementations:
  
Line 196: Line 304:
 
These examples demonstrate the versatility of chatbots across different industries and use cases. Chatbots continue to evolve with advancements in natural language processing and machine learning, providing more sophisticated and user-friendly interactions.
 
These examples demonstrate the versatility of chatbots across different industries and use cases. Chatbots continue to evolve with advancements in natural language processing and machine learning, providing more sophisticated and user-friendly interactions.
  
===Image recognition===
+
7. **Interactive Art: AI Dungeon**
Image recognition, a subset of computer vision, involves using artificial intelligence to interpret and understand the content of images. Here are some examples of image recognition implementations:
+
  - AI Dungeon is an example of an interactive storytelling application. It uses a generative language model to respond dynamically to user inputs, creating a unique and evolving narrative based on user choices. It demonstrates how generative AI can be applied to create interactive and personalized experiences.
  
1. **Facial Recognition:**
+
These examples showcase the diversity of generative AI applications across different domains. It's important to note that while generative AI has the potential for creativity and innovation, it also poses challenges related to ethics, bias, and the responsible use of technology.
  - Facial recognition systems can identify and verify individuals based on their facial features. These systems are used in various applications, including unlocking smartphones, secure access control systems, and law enforcement for identifying individuals in public spaces.
 
  
2. **Object Detection:**
+
===Accessibility enablers===
   - Object detection involves identifying and locating objects within an image. This technology is used in autonomous vehicles, surveillance systems, and robotics. For example, a security camera with object detection can identify and alert security personnel about suspicious objects or activities.
+
   - Operating systems, such as Windows, macOS, iOS, and Android, incorporate built-in speech synthesis features to assist users with visual impairments. These features read aloud text displayed on the screen, enabling visually impaired users to access information.
  
3. **Image Classification in Healthcare:**
+
Accessibility in Websites and Apps:**
   - Image recognition is widely used in healthcare for tasks such as identifying tumors in medical images (MRI, CT scans, etc.). Convolutional Neural Networks (CNNs) are often employed for image classification tasks in healthcare, helping radiologists and clinicians with diagnostics.
+
   - Websites and applications often integrate speech synthesis features to make content accessible to users with visual impairments or those who prefer audio content. This can include reading aloud articles, blog posts, or other textual information.
 
 
4. **Barcode and QR Code Scanning:**
 
  - Mobile applications often use image recognition to scan barcodes and QR codes. This is commonly used in retail for product identification, ticketing systems, and inventory management.
 
 
 
5. **Automated Optical Inspection (AOI) in Manufacturing:**
 
  - Image recognition is applied in manufacturing for quality control. Automated Optical Inspection (AOI) systems use image recognition to identify defects, anomalies, or inconsistencies in products during the production process.
 
 
 
6. **Satellite Image Analysis:**
 
  - Image recognition is utilized in analyzing satellite and aerial imagery. This application has uses in agriculture for crop monitoring, urban planning for infrastructure development, and environmental monitoring for assessing changes in ecosystems.
 
 
 
7. **Gesture Recognition:**
 
  - Gesture recognition involves interpreting human gestures captured by cameras. This technology is used in gaming consoles, smart TVs, and human-computer interaction systems. For example, a camera can recognize hand gestures to control a computer or gaming device.
 
 
 
8. **Retail Analytics:**
 
  - In retail, image recognition is used for customer analytics. Cameras can track customer movements and behaviors within a store, providing insights into customer preferences and helping retailers optimize store layouts and product placements.
 
 
 
9. **Security and Intrusion Detection:**
 
  - Image recognition is employed in security systems for identifying potential threats. This includes recognizing unauthorized individuals in secure areas or detecting unusual activities in public spaces.
 
 
 
10. **Social Media Image Tagging:**
 
    - Social media platforms use image recognition to automatically tag people in photos. The system can recognize faces and suggest tags, making it convenient for users to identify and share images with their friends.
 
 
 
11. **Wildlife Monitoring:**
 
    - Conservationists use image recognition to monitor wildlife in their natural habitats. Cameras equipped with image recognition technology can identify and track animals, helping researchers gather data on population dynamics and behavior.
 
 
 
These examples illustrate the wide range of applications for image recognition technology across different industries, highlighting its impact on automation, efficiency, and decision-making processes.
 
 
 
===Speech synthesis===
 
Speech synthesis, also known as text-to-speech (TTS), is a technology that converts written text into spoken words. Here are some examples of speech synthesis implementations:
 
 
 
1. **Accessibility Features:**
 
  - Operating systems, such as Windows, macOS, iOS, and Android, incorporate built-in speech synthesis features to assist users with visual impairments. These features read aloud text displayed on the screen, enabling visually impaired users to access information.
 
  
2. **Voice Assistants:**
+
===Voice assistants===
 
   - Voice assistants like Amazon Alexa, Google Assistant, and Apple's Siri use speech synthesis to respond to user queries. These systems can provide information, answer questions, and perform various tasks by converting text responses into natural-sounding speech.
 
   - Voice assistants like Amazon Alexa, Google Assistant, and Apple's Siri use speech synthesis to respond to user queries. These systems can provide information, answer questions, and perform various tasks by converting text responses into natural-sounding speech.
  
3. **Navigation Systems:**
 
 
   - GPS navigation systems use speech synthesis to provide turn-by-turn directions. The synthesized voice guides drivers and pedestrians, making it easier to navigate without having to look at a screen.
 
   - GPS navigation systems use speech synthesis to provide turn-by-turn directions. The synthesized voice guides drivers and pedestrians, making it easier to navigate without having to look at a screen.
  
4. **Interactive Voice Response (IVR) Systems:**
+
Interactive Voice Response (IVR) Systems
 
   - Many customer service and helpline systems use speech synthesis to interact with callers. These systems can provide information, guide users through menu options, and offer assistance without the need for a human operator.
 
   - Many customer service and helpline systems use speech synthesis to interact with callers. These systems can provide information, guide users through menu options, and offer assistance without the need for a human operator.
  
5. **E-learning Platforms:**
+
Call Center Automation:**
  - Speech synthesis is employed in e-learning platforms to provide narration for educational content. This enhances the learning experience by allowing users to listen to the content instead of reading it.
+
    - Speech synthesis is integrated into automated call center systems to deliver pre-recorded information or responses. This helps manage call volume and handle routine inquiries without the need for human operators.
  
6. **Accessibility in Websites and Apps:**
+
Voice Banking:**
  - Websites and applications often integrate speech synthesis features to make content accessible to users with visual impairments or those who prefer audio content. This can include reading aloud articles, blog posts, or other textual information.
+
    - Speech synthesis is utilized in voice banking applications that allow individuals to create personalized synthetic voices. This is particularly beneficial for people facing conditions that may impact their ability to speak, preserving their voice for future use.
  
7. **Language Translation Services:**
+
Smart Home Devices:**
  - Language translation services, such as Google Translate, use speech synthesis to convert translated text into spoken words. Users can listen to the translated content to better understand pronunciation and intonation.
 
 
 
8. **Entertainment and Media:**
 
  - Speech synthesis is used in the entertainment industry for various applications. For example, it can be used to create synthetic voices for characters in video games or to generate narration for audiobooks and podcasts.
 
 
 
9. **Smart Home Devices:**
 
 
   - Smart home devices, like smart speakers and connected appliances, use speech synthesis for communication. These devices can provide feedback, confirm actions, or deliver status updates using synthesized speech.
 
   - Smart home devices, like smart speakers and connected appliances, use speech synthesis for communication. These devices can provide feedback, confirm actions, or deliver status updates using synthesized speech.
  
10. **Voice Banking:**
+
===Language translators===
    - Speech synthesis is utilized in voice banking applications that allow individuals to create personalized synthetic voices. This is particularly beneficial for people facing conditions that may impact their ability to speak, preserving their voice for future use.
+
  - Language translation services, such as Google Translate, use speech synthesis to convert translated text into spoken words. Users can listen to the translated content to better understand pronunciation and intonation.
 
 
11. **Call Center Automation:**
 
    - Speech synthesis is integrated into automated call center systems to deliver pre-recorded information or responses. This helps manage call volume and handle routine inquiries without the need for human operators.
 
 
 
These examples demonstrate how speech synthesis technology enhances accessibility, user interaction, and the overall user experience across various domains and applications. Advances in natural language processing have led to more natural and expressive synthetic voices, making the technology increasingly sophisticated and user-friendly.
 
  
 
==AI Ethics==
 
==AI Ethics==
Line 284: Line 347:
  
 
Discuss and explain why it matters and how it can be addressed.
 
Discuss and explain why it matters and how it can be addressed.
 +
 +
==How ChatGPT works==
 +
: While being prompted, "Answer in a way that a kid can understand", an [[#AI system]] looks for those texts that are created by children or for children rather than is trying to find explicit instructions for that specific case.
 +
 +
ChatGPT, like other models in the GPT (Generative Pre-trained Transformer) family, is based on a transformer architecture. Here's a simplified overview of how ChatGPT works:
 +
 +
    Pre-training: ChatGPT is initially pre-trained on a large corpus of diverse text from the internet. During pre-training, the model learns to predict the next word in a sentence based on the context of the preceding words. This process helps the model learn grammar, facts, reasoning abilities, and even some level of world knowledge.
 +
 +
    Architecture: The underlying architecture of ChatGPT is the transformer architecture. Transformers use self-attention mechanisms to weigh the significance of different words in a sentence, allowing the model to capture long-range dependencies in the input data.
 +
 +
    Fine-tuning: After pre-training, the model can be fine-tuned on specific tasks or datasets. In the case of ChatGPT, it may be fine-tuned on conversational data to improve its performance in generating human-like responses in a chat setting.
 +
 +
    Prompt-based Interaction: When you interact with ChatGPT, you provide a prompt or a series of prompts, and the model generates a response based on its learned patterns from the pre-training data. The model doesn't have access to real-time data but relies on its pre-existing knowledge up to its last training cut-off.
 +
 +
    Response Generation: The model generates responses by sampling or selecting the most likely next word based on its learned probabilities. The generated responses aim to be contextually relevant and coherent based on the input prompt.
 +
 +
It's important to note that while ChatGPT can generate contextually relevant responses, it doesn't have a deep understanding or awareness of the world. It operates based on patterns learned during training and doesn't possess consciousness or true comprehension.
 +
 +
OpenAI may continue to update and improve models like ChatGPT over time, refining their capabilities and addressing limitations.
  
 
==Critical skills==
 
==Critical skills==
Line 514: Line 596:
  
 
Remember, the more specific and clear your prompts are, the more likely you are to receive helpful and relevant responses from ChatGPT.
 
Remember, the more specific and clear your prompts are, the more likely you are to receive helpful and relevant responses from ChatGPT.
 +
 +
==Prompt generators==
 +
ChatGPT prompt generators are tools that help you create effective and engaging prompts for ChatGPT, a conversational AI model based on OpenAI’s GPT-4. ChatGPT prompt generators can help you customize your interaction with ChatGPT by setting a specific role, tone, style, and objective for the response.
 +
 +
Some of the best ChatGPT prompt generators that I found on the web are:
 +
 +
ChatGPT Prompt Generator | Welcome AI: This tool lets you experiment with different prompts for ChatGPT by providing some context and your goal or objective. It also follows OpenAI’s best practices for prompt design.
 +
ChatGPT Prompt Generator | Prompt Advance: This tool allows you to enforce the style of the response by choosing from a list of predefined roles, such as friend, teacher, coach, etc. It also lets you copy the prompt with one click.
 +
AI ChatGPT Prompt Generator | Taskade: This tool uses OpenAI’s GPT-4 model to generate engaging prompts based on your specified inputs. You can also adjust the temperature and top-k parameters to control the randomness and diversity of the response.
  
 
==Prompt parameters==
 
==Prompt parameters==
Line 531: Line 622:
  
 
It's important to note that while ChatGPT can understand and respond to a wide range of prompts, the clarity and specificity of your input can significantly influence the quality of the generated output. Experimenting with different parameters and refining your prompts can help you get the desired results.
 
It's important to note that while ChatGPT can understand and respond to a wide range of prompts, the clarity and specificity of your input can significantly influence the quality of the generated output. Experimenting with different parameters and refining your prompts can help you get the desired results.
 
==Fine tuning==
 
 
As of my last knowledge update in January 2022, OpenAI has not released a fine-tuning option for GPT-3, including ChatGPT. Fine-tuning typically involves training a pre-trained model on a specific dataset for a more narrow and task-specific purpose. OpenAI has not provided the ability for users to fine-tune GPT-3 models, likely due to concerns related to misuse and the potential for biases in the fine-tuned models.
 
 
As policies and features may change, I recommend checking the official OpenAI documentation or website for the most up-to-date information on model capabilities and any new features that may have been introduced since my last update.
 
 
If fine-tuning becomes available for ChatGPT or other models in the future, OpenAI would likely provide guidelines and documentation on how users can perform fine-tuning. Always make sure to follow OpenAI's terms of service and guidelines for ethical and responsible use of their models.
 
  
 
==Prompt techniques==
 
==Prompt techniques==
Line 575: Line 658:
 
```
 
```
 
These strategies should generally work well across different conversational AI models built on top of the Hugging Face Transformers library, including HuggingChat. Adapt them according to your particular needs to optimize interactions with the model.
 
These strategies should generally work well across different conversational AI models built on top of the Hugging Face Transformers library, including HuggingChat. Adapt them according to your particular needs to optimize interactions with the model.
 +
 +
==Parameters==
 +
ChatGPT prompts can have various parameters that affect the quality and style of the generated responses. Some of the common parameters are:
 +
 +
length: This parameter controls the length of the response, measured in number of tokens or words. You can adjust this value to ensure the output is neither too short nor too lengthy.
 +
top_p: This parameter controls the diversity of the response, by filtering out low-probability words. Higher values lead to more diverse responses, while lower values generate more conservative and predictable responses.
 +
frequency_penalty: This parameter affects the likelihood of generating words that have already been used in the response. Higher values result in more varied responses, while lower values lead to more repetitive responses.
 +
presence_penalty: This parameter penalizes the likelihood of generating words that are not present in the input prompt. Higher values result in more relevant responses, while lower values lead to more irrelevant responses.
 +
stop_sequence: This parameter specifies a sequence of words that ChatGPT should avoid generating in its response. You can use this parameter to prevent the generation of inappropriate or sensitive content, or to maintain the focus of the conversation.
 +
You can learn more about these and other parameters from these sources:
 +
 +
Mastering ChatGPT Prompts: A Guide to Using Parameters by Bruce Lim
 +
How to Prompt ChatGPT - A Helpful Guide by allPrompts
 +
How To Write ChatGPT Prompts: Your 2024 Guide by Coursera
 +
195 ChatGPT Prompts (& How to Write Your Own) by Semrush
  
 
==Conclusion==
 
==Conclusion==
Line 585: Line 683:
 
(4) Introduction to Artificial Intelligence (AI) | Coursera. https://www.coursera.org/learn/introduction-to-ai.
 
(4) Introduction to Artificial Intelligence (AI) | Coursera. https://www.coursera.org/learn/introduction-to-ai.
 
(5) How to Learn Artificial Intelligence: A Beginner’s Guide. https://www.coursera.org/articles/how-to-learn-artificial-intelligence.
 
(5) How to Learn Artificial Intelligence: A Beginner’s Guide. https://www.coursera.org/articles/how-to-learn-artificial-intelligence.
 +
 +
https://github.com/f/awesome-chatgpt-prompts
 +
https://promptbase.com/
 +
 +
==Key terms==
 +
===AI===
 +
===AI system===
 +
===Inference===
 +
: Once trained, the model uses an inference algorithm to make predictions on new, unseen data. This involves applying the learned patterns and relationships to generate output based on input features.
 +
 +
: The inference algorithms, particularly, are used to process data and perform tasks that [[#End-user]]s prompt.
 +
 +
===Intelligent machine===
 +
An intelligent machine refers to a device or system that possesses the ability to perform tasks that typically require human intelligence. This includes capabilities such as learning from experience, adapting to new situations, understanding natural language, solving complex problems, and making decisions.
 +
 +
Intelligent machines can take various forms, ranging from simple algorithms and rule-based systems to advanced artificial intelligence (AI) systems that utilize machine learning, deep learning, and other sophisticated techniques. These machines can be designed to perform specific tasks, such as image recognition, language translation, playing strategic games, or even driving vehicles.
 +
 +
The level of intelligence in a machine can vary, and it's often categorized into narrow or weak AI, which is designed for a specific task, and general or strong AI, which would have the ability to perform any intellectual task that a human being can. As of my last knowledge update in January 2022, we have achieved significant progress in narrow AI applications, but the development of general AI remains a complex and ongoing challenge.
 +
 +
===ML===
 +
===Training infrastructure===
 +
: Efficient hardware and software infrastructure are necessary for training large-scale AI models. Without the appropriate infrastructure, training times can be prohibitively long, hindering the development process.
 +
 +
===Deployment infrastructure===
 +
: Once trained, [[#AI model]]s need a robust deployment infrastructure to function in real-world applications. This includes considerations for scalability, reliability, and integration with existing systems.
 +
 +
==See also==
 +
* https://ainurtz.com/
 +
* https://futureofworkisai.com/ -- links, logo, done
 +
* https://hotcoe.com/
 +
* https://howsafeismyjob.com/
 +
* https://in2job.com/
 +
* https://ismyjobsecure.com/
 +
* https://ksacerts.com/
 +
* https://plzin.com/
 +
* https://wisenxt.com/

Latest revision as of 05:41, 3 February 2024

Contents

What AI is about

What AI is

Non-intelligent machines simply execute strict commands of their operators. Non-intelligent software follow explicit rules programmed by human developers.
#AI stands for #Artificial intelligence, which is the competency of an artificial system to learn and perform partially or fully autonomously based on what this system such as a computer, drone, self-driving car, or robot has learned.
For instance, Google-owned AlphaZero is given no knowledge beyond basic chess rules. Instead, this #AI model is prompted to play against itself and, using #Reinforcement learning, to learn from its own mistakes. Win or lose, AlphaZero piles up lessons learned from its millions and millions of games. According to its creator, Demis Hassabis,
"Actually you set off AlphaZero in the morning and it starts off playing randomly. By lunchtime, you know it's able to beat [...] most chess players. And then, by the evening, it's stronger than the world champion."
Because of their #Computing power and #Scalable persistent memory, #AI systems can learn faster and accumulate more knowledge than any human being. However, they are able to learn only in those domains and in those ways that their creators specify. No #Intelligent machine of today can learn without #Algorithms.

What AI is not

#AI-powered systems may mimic or, more exactly, replicate human intelligence. However, they are not beings, individuals, personages, nor souls. They are tools.
#Intelligent machines of today lack a true understanding of its surroundings and, overall, the world at least yet. For example, #AI-driven systems can sound like humans because, similarly to parrots, they learn from human speech samples.
No single #AI system possesses consciousness, intentions, interests, motivations, subjective experiences, self-awareness or awareness of their existence. Similarly to "special police" not being regular police, as of now, artificial intelligence is not general intelligence at least yet.

AI vs ML

#ML is a separate area of studies within #AI. However, no #AI can be possible without #ML. When people say, "AI and ML", they may be emphasizing the distinction between traditional AI methods and approaches that involve machine learning techniques.
Traditional #AI includes rule-based systems and expert systems, while #Machine learning relies on data-driven approaches to improve performance over time. By mentioning both ML and AI, individuals may want to acknowledge the broader context of AI and highlight the specific role that machine learning plays within that framework.

Animals vs AI

Naturally, intelligence is a capacity of animals, especially of humans. Some argue that plants demonstrate limited intelligence as well. That is why the intelligence that computers or robots may demonstrate is called "artificial".
Parents teach their kids, while giving them initial instructions and exposing them to the real world to learn. To train an #AI system, their creators supply this system with #Data and #Algorithms to process it. No parents can predict everything their kids do; similarly, designers of #AI models cannot anticipate everything their creations would produce.
Nevertheless, there is a distinctive difference between raising children and training of machines. Parental controls over what their kids learn and over patterns of their learning are limited. Developers of #AI models today have full control over #Algorithms that the machines utilize to learn and perform, as well as over #Data provided during pre-training.

Capability of AI

What AI can

By definition, #AI can perform at least one task autonomously, where "autonomously" refers to the limited ability of an #AI system to function independently without external control or direct human intervention.
#AI systems have the capability to execute tasks that are not exclusively programmed in advance. They can be grouped in three task sets:
  1. Partially-autonomous acquisition of data through listening, reading, and graphic recognition. These tasks are commonly known as #Machine learning (#ML).
  2. Partially-autonomous discrimination of data through its analysis and classification. The word "discrimination" originates from the Latin verb "discriminare," which means "to distinguish" or "to separate." Those #AI models that perform these tasks are commonly known as #Discriminative models.
  3. Partially-autonomous generation of results through delivery of produced outputs such as documents and commands, for instance, needed for autonomous driving. Those #AI models that perform these tasks are commonly known as #Generative models.
Generally speaking, autonomy implies some level of self-governance and the ability to adapt to changing circumstances without relying on continuous commands, guidance, or instructions from external sources.

What AI can't yet

As of today, #AI systems are only partially autonomous, lacking the ability to act beyond their specified functions and requiring human guidance for contextual understanding, goal-setting, and supervision. Currently, #AI:
  • Can't perform without #Algorithms and, often, training: #AI models encounter challenges when adapting to unseen contexts due to their reliance on the training data, struggling with entirely new or unforeseen situations that were not part of their training set and lacking the ability to generalize knowledge like humans. While AI can generate novel outputs based on learned patterns, it falls short of true creative thinking, imagination, and inspiration, as it cannot form unique concepts beyond the confines of its training data.
  • Have no common or conventional senses: AI lacks inherent ethics. Ethical concerns arise from human developers' decisions and biases in training data, unintentionally perpetuating biases. Despite potency in tasks, AI faces challenges in common sense reasoning, interpreting implicit meanings, grasping context, and making inferences beyond training data.
  • Have no emotion, intuition, or feeling: AI lacks the capacity for intuition or gut feelings, as human intuition often involves subconscious processing and experiences that are difficult to reproduce in machines. While certain AI applications may simulate emotional responses or recognize human emotions, such simulations are rooted in pattern recognition and do not constitute a true understanding of emotions.
As of today, #AI systems operate within the parameters set by their programming with #Algorithms and training.

AI vs AGI

As of now, #AI refers to limited intellectual capacity called #Narrow AI. #Artificial intelligence of today can perform partially autonomously narrow, specific tasks such as chatbots, facial and speech recognition, recommendation systems, self-driving cars, and virtual assistants.
On the contrary, #Artificial general intelligence or #AGI, if developed, would be able to perform any intellectual task that a human can do. #General AI is in the research phase and is not commercially available.
In 2019, Elon Musk expressed his belief that #AGI will be possible by 2029. In the same year, China-based AI visionary Kai-Fu Lee said in an interview that #AGI would be developed, "not within the next 30 years and possibly never".
If researchers are able to develop an AGI system, it could have significant implications for society, including creation of machines that are more intelligent than humans, which could pose ethical and safety concerns.

Growth of AI capacity

Impact on jobs

How Safe Is My Job

AI system

An #AI system refers to a #System such as a computer, drone, self-driving car, or robot that can perform one or more tasks autonomously, without explicit human command. Some people refer #AI systems to software only and use terms such as #AI-driven systems, #AI-powered systems, or #Intelligent machines to indicate not-only-software systems. However, those terms are often used interchangeably.

Literally, #AI is software. Any software requires hardware, which stands for physical devices that allows for software to run. Any, robotic or non-robotic, #AI system needs (a) #Computing power to compute, (b) #Persistent memory to store #Data, (c) actuators, dashboards, keyboards, sensors, or other #AI peripherals, as well as (d) connectors and wires to connect the system's components.

Besides #AI models and hardware, #AI needs #AI agents, #Algorithms, #Data, and the software to enable those. So, when we say #AI, we often refer to some combination of software and hardware pieces. Depending on the type and complexity of the task, #AI systems can use different combinations of hardware and software components.

AI agent

An AI agent is any software or hardware component of an #AI system that is responsible for:
  1. Input data intake through perceiving the system's environment. To collect data, AI agents often have #AI peripherals.
  2. System's decision through using algorithms, rules, or machine learning models to analyze the information they receive and make decisions based on that analysis. The decision-making process is typically aimed at achieving a specific goal or solving a particular problem.
  3. Output delivery such as taking actions to achieve specific goals. Once an #AI system made a decision, the AI agent takes actions in the environment. This could involve physical actions in the case of a robotic agent or virtual actions in the case of software-based agents.
Based on their level of autonomy and learning capability, AI agents may be divided in (a) reactive agents that follow predefined rules and respond to specific situations without learning from experience and (b) intelligent agents that are capable to learn from experience, adapt their behavior based on feedback, improve performance over time, and/or acquire new knowledge to handle different situations. Intelligent agents are #AI systems on their own.

Algorithms

In the context of #AI systems, #Algorithms are the rules or instructions that #AI systems follows to complete their tasks. In the broader sense, an algorithm is a step-by-step set of directives, procedures, and/or regulations that dictate how a specific task or computation should be performed.
#AI systems normally utilize several #Algorithms. Generally speaking, the more complex #AI system is, the more #Algorithms it tends to utilize. The #Algorithms can be divided in two categories:
  1. Those that are used in #Model development.
  2. Those that are used in #Inference or making predictions on new, unseen data. The #Inference algorithms, particularly, are used to process data and perform tasks that #End-users prompt.
#Algorithms can be based on logic, optimization, or statistical techniques. Data scientists and engineers who specialize in #AI and #ML design mathematical models, develop learning algorithms, implement them using programming languages, test their predictions, and adjust them based on the testing.

Computing power

In the context of #AI systems, #Computing power refers to the capabilities of the hardware processors that perform the tasks and computations associated with artificial intelligence (AI) applications.
  • #Central processing units (#CPUs) are normally used in simple models on small datasets, inference on low-resource devices, non-deep learning algorithms, and demonstrations. Because of their limited number of cores, they are not sufficient in running more complex computations in #AI.
  • #Graphics processing units (#GPUs) offer significant advantages in terms of accelerating various AI processes in comparison with #CPUs. #GPUs have hundreds or even thousands of cores, which allow them processing their tasks simultaneously in parallel. Although #GPUs were initially designed for graphics processing, they make early #Deep learning possible. Large-scale models with millions or billions of parameters require more powerful hardware accelerators than #CPUs.
  • #Tensor Processing Units (#TPUs) are the newest category of processors in #AI. Google developed them specifically for #Deep learning. However, because of #TPUs' significant costs, they have been only used for some of the most recent and advanced models, such as GPT-4 and CodeLlama.
As of early 2024, the statistics is not available, but it seems that #GPUs are still the most widely used hardware processors for AI computing, followed by #TPUs, which are gaining popularity and adoption. #CPUs are likely to have a very low percentage of AI computing, as they are not efficient or powerful enough for most AI tasks.
In the meanwhile, once trained, any #AI model needs a #Deployment infrastructure to function in real-world applications. Efficient hardware and software #Training infrastructure is necessary for training large-scale AI models. Without the appropriate infrastructure, training times can be prohibitively long, hindering the development process.

Data

In the context of #AI systems, #Data is the raw material that #AI systems use to learn from and make predictions. #Data can be #Structured data (such as categories, labels, or numbers) or #Unstructured data (such as audio, images, or text). #End-user's input such as a #Prompt is a part of #Data as well.

#Persistent memory

Scalable persistent memory

Human memory is not a persistent storage system and is subject to decay and changes over time.

Data intake and storage

Including code, image, text, video, and voice acquisition. AI systems can accept large amounts of data faster and more accurately than humans. For example, AI can identify toys, faces, animals, and even diseases from images. They can also understand and transcribe human speech with high accuracy.

AI model

An #AI model is the knowledge abstraction or representation that makes an #AI system intelligent.

own knowledge abstractions called #AI models.


#Model development

Model architecture

#AI systems often deploy multiple models and their combination is called #Model architecture.
Different models make predictions, classifications, or decisions based on new or unseen data. Some models work with input agents, some do with decision agents, and same work with output agents. For instance, in #GPT,
  • A dedicated tokenization model chunks #End-user's request in smaller tokens.
  • #Discriminative models process the chunked tokens. One model maps it to a high-dimensional vector, known as a token embedding. Another model adds to this embedding token's position in a prompt, etc.
  • Finally, #Generative models are used to generate the output. For instance, one model may suggest the output format, another may so its content, the third may make sure that the output fits the audience, etc.
Any #AI model generalizes its learning from the training data to perform tasks on new, similar data. That way, the models can capture patterns, relationships, or features that are relevant for a specific task or domain.

Large language model

Workflow of AI system

For example, when a user prompts a #Generative Pre-trained Transformer or #GPT to answer his or her question, this AI system:
  1. Transforms the unstructured human prompt, while encoding it into a set of structured machine requests. The machine requests identify the prompted answer format, the topic or topics to address, etc. To create structured requests, the system compares the human request with the prompt dataset.
  2. Studies those areas of its large dataset that are relevant to the request. To create that dataset, the system receives the data, usually, an enormous volume of data, in advance. This process is known as pre-training.
  3. Generates the response, while selecting most likely words in a most likely sequence from large dataset.

Model development

#Algorithms, both learning and optimization, are integral to creating models in machine learning. Learning algorithms such as ones for #Training, #Evaluation, and/or #Hyperparameter tuning enable models to discover patterns in data. #Optimization algorithms adjust model parameters for optimal performance during training.

Training

During the training phase, an algorithm is employed to optimize the parameters of a model. This involves presenting the model with a dataset, allowing it to learn patterns and relationships within the data. The algorithm adjusts the model's parameters iteratively to minimize the difference between its predictions and the actual outcomes in the training data.

Optimization

Optimization algorithms are often used to fine-tune the parameters of a model to improve its performance. Gradient descent is a common optimization algorithm that adjusts the model's parameters based on the gradient of the loss function with respect to those parameters.

Evaluation

Algorithms are also used to evaluate the performance of models. Metrics such as accuracy, precision, recall, and F1 score are calculated using specific algorithms to assess how well the model generalizes to new data.

Hyperparameter tuning

Algorithms play a role in the process of hyperparameter tuning, where the values of hyperparameters (settings not learned during training) are adjusted to find the optimal configuration for the model.

AI peripheral

In the context of #AI systems, peripheral devices or #AI peripherals enable the system's interaction with the physical world. The peripherals represent the system's hardware beyond their core computing. To interact with the digital world, #AI systems utilize #APIs.

The peripheral devices interface with or connect to #AI agents to facilitate their communication with their their physical environment. End-users of #AI systems use the peripherals to work with these systems.

Input devices

AI input devices enable input from the physical world into #AI systems.
  • Manipulators such as buttons, joysticks, keyboards, mouses, steering wheels, etc. enable human manual input into #AI systems. Utilizing those, #End-users may manually input commands, queries, or parameters.
  • Sensors such as accelerometers, cameras, microphones, radars, scanners, and etc. enable automatic input into #AI systems.

Output devices

AI output devices enable output from #AI systems into the physical world. To impact the real world, a few devices can be used such as:
  • Actuators prompt physical actions of motors, robotic arms, or other mechanisms when #AI systems make such decisions.
  • Dashboards and displays (except #Two-way portals) present or display relevant information or insights. Dashboards visually represent data and information in a user-friendly way that allows users to monitor and analyze key metrics or performance indicators. They provide a centralized and easily digestible view of information. Dashboards and displays are not inherently designed for two-way communication, direct interaction, or data input from users. Touch-screen displays and similar devices belong to #Two-way portals.
  • Printers: Output devices that produce tangible results based on the AI system's outputs. This could include generating reports, images, or other forms of information.

Two-way portals

Two-way portals are interactive platforms where #End-users can both consume and contribute content. While utilizing interactive websites, collaboration platforms, and customer portals, the users can not only view information, but also actively participate by submitting data, making updates, or engaging in communication.

Key concepts of AI

AI concepts overlap; advancements in one area can benefit others. AI is a rapidly evolving field, and new concepts and interdisciplinary approaches continue to emerge.

Computer vision

#Computer vision involves the use of algorithms and artificial intelligence to enable computers to interpret and understand visual information from the world, typically through images or videos. It can be employed for various tasks, including object recognition, image classification, facial recognition, and even autonomous navigation for vehicles.

Deep learning

#Deep learning is a subset of machine learning that involves neural networks with multiple layers (deep neural networks) to automatically learn hierarchical representations of data. It excels in tasks such as image and speech recognition, natural language processing, and complex pattern recognition, enabling systems to automatically extract intricate features and make high-level abstractions from raw input data.

Expert systems

#Expert systems emulate human expertise by using a knowledge base and a set of rules to make decisions or solve problems within a specific domain. They can be applied in areas such as diagnosis, decision-making, and troubleshooting, providing intelligent solutions by mimicking the decision-making process of human experts.

Fine-tuning

#Fine-tuning in machine learning involves adjusting pre-trained models to better suit specific tasks or datasets, leveraging existing knowledge for improved performance in specialized applications such as image classification or natural language processing. It allows for optimization without starting the training process from scratch, making it a valuable technique for enhancing model efficiency and effectiveness.

Machine learning

#Machine learning is a subset of artificial intelligence that involves the use of algorithms to enable computers to learn patterns and make predictions or decisions from data without explicit programming. It can be applied to various tasks, including image recognition, recommendation systems, and predictive analytics, allowing systems to improve their performance over time through experience with new information.

Machine learning is a specific approach within the field of AI that involves creating algorithms and models that enable machines to learn patterns and make decisions without being explicitly programmed.

Natural language processing

#Natural language processing (#NLP) involves the use of computational algorithms to analyze and understand human language, allowing computers to interpret, generate, and respond to text or speech. NLP can be applied to tasks such as sentiment analysis, language translation, and chatbot development, enabling machines to interact with and comprehend human communication.

Neural networks

#Neural networks are computational models inspired by the human brain's structure, composed of interconnected nodes organized into layers. They can learn complex patterns from data, making them versatile for tasks such as image and speech recognition, natural language processing, and solving intricate problems across various domains.

Reinforcement learning

#Reinforcement learning is a machine learning paradigm where an agent learns to make decisions by interacting with an environment and receiving feedback in the form of rewards or penalties. It can be applied to tasks such as game playing, robotic control, and optimization problems, enabling systems to learn optimal strategies through trial and error.

Robotics

#Robotics involves the design, creation, and operation of robots, which are autonomous or semi-autonomous machines capable of performing tasks in the physical world. Robotics can be applied in various fields, including manufacturing, healthcare, exploration, and logistics, to automate processes, assist humans, and execute complex operations in environments that may be hazardous or impractical for humans.

Supervised learning

#Supervised learning is a machine learning paradigm where the algorithm is trained on a labeled dataset, meaning it learns from input-output pairs to make predictions or classifications on new, unseen data. It can be applied to tasks such as spam detection, image recognition, and speech recognition, where the algorithm generalizes patterns from known examples to make accurate predictions on new instances.

Discriminative AI

#Discriminative AI refers to #Discriminative models. They are designed to (a) classify input data and/or (b) predict an outcome based on input data.

Acknowledged speech

#Discriminative models can acknowledge spoken language and convert it into text. Speech recognition systems learn to map audio features to words or phrases, making them useful in applications like virtual assistants and transcription services.

Classified code or text

#Discriminative models can categorize topics, classify scripts, both code and text, detect spam, and analyze sentiments in #Natural language processing (#NLP).

Detected activity

#Discriminative models can detect activities in cybersecurity, finance, and retail by analyzing patterns, anomalies, and trends in data to identify and categorize different cyber threats, financial transactions, and consumer behaviors. To detect fraud, these models learn patterns of normal behavior and can identify deviations that may indicate fraudulent transactions.

Diagnosed condition

#Discriminative models can classify symptoms, diagnose medical conditions, and predict patient outcomes based on medical images and other patient's data. #Convolutional neural networks (#CNNs) are often employed for classification tasks performed on #Recognized images in healthcare, helping radiologists and clinicians with diagnostics.

Recognized image

#Image recognition, a subset of #Computer vision, involves using artificial intelligence to interpret and understand the content of images. Besides recognizing faces, gestures, and illustrations, image recognition can be used to:
  • Analyze satellite image in agriculture for crop monitoring, urban planning for infrastructure development, and environmental monitoring for assessing changes in ecosystems.
  • Convert handwritten characters or words into digital formats.
  • Detect unusual activities in finance and security, irregular changes, unauthorized objects in secure areas, tumors in medical images such as MRI, CT scans, etc..
  • Identify defects, anomalies, or inconsistencies in products either in manufacturing or maintenance via automated optical inspection (AOI).
  • Scan barcodes and QR codes. This is commonly used in retail for product identification, ticketing systems, and inventory management.
  • Tag people automatically in social media photos.
  • Track customer movements and behaviors within a store, providing insights into customer preferences and helping retailers optimize store layouts and product placements.
Image recognition often include #Object detection as a specific task within its scope; while image recognition generally refers to identifying and categorizing objects or patterns in images, object detection specifically involves locating and delineating individual objects within an image. Obviously, before an object can be recognized, it should be detected.

Generative AI

#Generative AI refers to #Generative models that can create new content, script, etc. Two primary approaches are the following.

New object synthesis

AI-powered new object synthesis involves generating new content, text, code, image, etc. from scratch in order to create #Synthesized objects.

Feature transfer

AI-powered feature transfer involves application of existing features of one objects to different objects in the context of generative AI in order to apply #Transferred features.

Transferred features

AI-powered feature transfer refers to the process of applying the behaviors, style, nonverbal communications, and/or other features of one object to another while preserving the content of the second object. In feature transfer, the model extracts specified features from one object and applies them to another.

This process can also be called "neural feature transfer." The terms such as "style transfer", "nonverbal cues transfer", etc. can describe specific processes.

Deepfake

#Deepfakes are realistic forgeries combining #Generated image, #Generated video, and/or #Synthesized speech. StyleGAN (Generative Adversarial Network) is one of the leading companies that specialize in those applications.

Transferred artistic style

#Neural style transfer involves using a pre-trained convolutional neural network to blend the content of one image with the style of another. Challenges include computational costs, potential artifacts in generated images, and the need for careful parameter tuning to achieve visually pleasing results.

Transferred nonverbal cues

#Deep Video Portraits is a technique that enables photo-realistic re-animation of portrait videos using only an input video. It manipulates or synthesizes video content with a focus on accurately transferring facial expressions and movements from one person to another in a video. It has applications in video editing, movie post-production, visual effects, visual dubbing, virtual reality, and telepresence.

Synthesized objects

AI-powered synthesis involves creating entirely new code, image, text, sound, video, or even entire pieces of art from scratch. In this case, "scratch" refers to the #AI models that are created as a result of #Training.

Challenges in AI generation encompass avoiding inaccuracies, biases, and ethical concerns, whether in text, images, videos, music, or speech. Key hurdles include ensuring realism, coherence, and addressing potential biases in training data for responsible and inclusive generation across diverse domains.

Generated code or text

AI script generators are constructed using deep learning architectures like #Recurrent neural networks (#RNNs) or transformers, trained on diverse code and/or textual datasets. As of early 2024, the leading script generators are OpenAI's GPT Models (ChatGPT-3 and ChatGPT-4), Google's Bard, and HummingFaces.

Generated image

AI image generators are typically constructed using deep neural networks, such as #Generative adversarial networks (#GANs) or #Variational autoencoders (#VAEs), trained on extensive datasets of images. Some image generators are designed to generate images from textual descriptions. For instance, you could ask DALL-E to generate images of "a one-story blue house shaped like a flower." As of early 2024, the leading image generators are OpenAI's DALL-E, Google's DeepDream, and Midjourney.

Generated video

AI video generators are built using deep learning techniques, particularly through the utilization of #Generative models such as #Generative adversarial networks (#GANs) or #Variational autoencoders (#VAEs) trained on large datasets of videos.

Synthesized music

AI music generators are built using machine learning algorithms, often employing techniques such as Recurrent neural networks (#RNNs) or transformer models trained on vast datasets of musical compositions. As of early 2024, the leading music generator is OpenAI's MuseNet.

Synthesized speech

Speech synthesis, also known as #Text-to-speech (#TTS), is a technology that converts written text into spoken words. AI speech generators are built using deep learning techniques, particularly with #Recurrent neural networks (#RNNs) or #Transformer models, trained on large datasets of human speech.

Assistants and chatbots

Chatbots are applications that use natural language processing (NLP) and machine learning to engage in conversations with users. They find applications in customer support, information retrieval, and various other domains. Here are a few examples of chatbot implementations:

1. **Customer Support Chatbots:**

  - Many companies use chatbots to provide instant support to customers on their websites. These chatbots can answer frequently asked questions, troubleshoot common issues, and guide users through various processes. For example, a chatbot on an e-commerce site might help users track orders or find product information.

2. **Virtual Assistants:**

  - Virtual assistants like Apple's Siri, Google Assistant, and Amazon's Alexa are examples of chatbots that assist users with tasks and answer questions. They can perform actions based on voice commands or text inputs, such as setting reminders, sending messages, or providing weather updates.

3. **Healthcare Chatbots:**

  - Healthcare chatbots can assist users in finding information about symptoms, scheduling appointments, or understanding medication instructions. They are designed to provide preliminary information and guidance. For instance, a healthcare chatbot might ask about symptoms and provide general advice based on the user's responses.

4. **Education and Training:**

  - Chatbots can be used in educational settings to provide information, answer questions, and even assist with learning exercises. They can act as virtual tutors or help users navigate educational resources. Duolingo, for example, uses a chatbot to simulate conversations for language learners.

5. **HR and Recruitment Chatbots:**

  - Chatbots can streamline the recruitment process by engaging with job candidates, answering queries about job openings, and collecting initial information. They can also assist employees with HR-related inquiries. For example, a chatbot might help employees understand company policies or request time off.

6. **Finance and Banking Chatbots:**

  - Chatbots in the financial industry can assist users with tasks like checking account balances, transferring funds, or getting information about recent transactions. They can also provide financial advice or answer queries about banking services.

7. **Social Media Chatbots:**

  - Some businesses use chatbots on social media platforms to engage with customers. These chatbots can provide information about products, assist with purchases, or address common inquiries. Facebook Messenger, for example, supports chatbots for business pages.

8. **Entertainment and Gaming:**

  - Chatbots are used in gaming and entertainment platforms to enhance user experience. They can provide game-related information, tips, and even engage in interactive storytelling. Additionally, chatbots can simulate conversations with fictional characters.

9. **Legal Assistance Chatbots:**

  - Chatbots can provide basic legal information, guidance on legal processes, and help users understand their rights. They are not a substitute for legal professionals but can offer preliminary assistance. For instance, a legal chatbot might help users generate simple legal documents.

These examples demonstrate the versatility of chatbots across different industries and use cases. Chatbots continue to evolve with advancements in natural language processing and machine learning, providing more sophisticated and user-friendly interactions.

7. **Interactive Art: AI Dungeon**

  - AI Dungeon is an example of an interactive storytelling application. It uses a generative language model to respond dynamically to user inputs, creating a unique and evolving narrative based on user choices. It demonstrates how generative AI can be applied to create interactive and personalized experiences.

These examples showcase the diversity of generative AI applications across different domains. It's important to note that while generative AI has the potential for creativity and innovation, it also poses challenges related to ethics, bias, and the responsible use of technology.

Accessibility enablers

  - Operating systems, such as Windows, macOS, iOS, and Android, incorporate built-in speech synthesis features to assist users with visual impairments. These features read aloud text displayed on the screen, enabling visually impaired users to access information.

Accessibility in Websites and Apps:**

  - Websites and applications often integrate speech synthesis features to make content accessible to users with visual impairments or those who prefer audio content. This can include reading aloud articles, blog posts, or other textual information.

Voice assistants

  - Voice assistants like Amazon Alexa, Google Assistant, and Apple's Siri use speech synthesis to respond to user queries. These systems can provide information, answer questions, and perform various tasks by converting text responses into natural-sounding speech.
  - GPS navigation systems use speech synthesis to provide turn-by-turn directions. The synthesized voice guides drivers and pedestrians, making it easier to navigate without having to look at a screen.

Interactive Voice Response (IVR) Systems

  - Many customer service and helpline systems use speech synthesis to interact with callers. These systems can provide information, guide users through menu options, and offer assistance without the need for a human operator.

Call Center Automation:**

   - Speech synthesis is integrated into automated call center systems to deliver pre-recorded information or responses. This helps manage call volume and handle routine inquiries without the need for human operators.

Voice Banking:**

   - Speech synthesis is utilized in voice banking applications that allow individuals to create personalized synthetic voices. This is particularly beneficial for people facing conditions that may impact their ability to speak, preserving their voice for future use.

Smart Home Devices:**

  - Smart home devices, like smart speakers and connected appliances, use speech synthesis for communication. These devices can provide feedback, confirm actions, or deliver status updates using synthesized speech.

Language translators

  - Language translation services, such as Google Translate, use speech synthesis to convert translated text into spoken words. Users can listen to the translated content to better understand pronunciation and intonation.

AI Ethics

ethical issues related to AI

Bias

Discuss and explain why it matters and how it can be addressed.

Privacy

Discuss and explain why it matters and how it can be addressed.

Accountability

Discuss and explain why it matters and how it can be addressed.

Transparency

Discuss and explain why it matters and how it can be addressed.

How ChatGPT works

While being prompted, "Answer in a way that a kid can understand", an #AI system looks for those texts that are created by children or for children rather than is trying to find explicit instructions for that specific case.

ChatGPT, like other models in the GPT (Generative Pre-trained Transformer) family, is based on a transformer architecture. Here's a simplified overview of how ChatGPT works:

   Pre-training: ChatGPT is initially pre-trained on a large corpus of diverse text from the internet. During pre-training, the model learns to predict the next word in a sentence based on the context of the preceding words. This process helps the model learn grammar, facts, reasoning abilities, and even some level of world knowledge.
   Architecture: The underlying architecture of ChatGPT is the transformer architecture. Transformers use self-attention mechanisms to weigh the significance of different words in a sentence, allowing the model to capture long-range dependencies in the input data.
   Fine-tuning: After pre-training, the model can be fine-tuned on specific tasks or datasets. In the case of ChatGPT, it may be fine-tuned on conversational data to improve its performance in generating human-like responses in a chat setting.
   Prompt-based Interaction: When you interact with ChatGPT, you provide a prompt or a series of prompts, and the model generates a response based on its learned patterns from the pre-training data. The model doesn't have access to real-time data but relies on its pre-existing knowledge up to its last training cut-off.
   Response Generation: The model generates responses by sampling or selecting the most likely next word based on its learned probabilities. The generated responses aim to be contextually relevant and coherent based on the input prompt.

It's important to note that while ChatGPT can generate contextually relevant responses, it doesn't have a deep understanding or awareness of the world. It operates based on patterns learned during training and doesn't possess consciousness or true comprehension.

OpenAI may continue to update and improve models like ChatGPT over time, refining their capabilities and addressing limitations.

Critical skills

Using ChatGPT effectively requires a set of critical skills to maximize the value and obtain accurate, relevant, and meaningful information. Here are some skills and considerations for a ChatGPT user:

1. **Clear Communication Skills:**

  - Articulate your questions or prompts clearly and concisely.
  - Use proper grammar and syntax to enhance understanding.

2. **Domain Knowledge:**

  - Possess knowledge about the subject matter you're discussing to guide the conversation effectively.

3. **Context Management:**

  - Provide context for your questions to help ChatGPT understand the specifics of your inquiry.
  - Refer back to previous messages to maintain context and coherence.

4. **Prompt Iteration:**

  - Be prepared to iterate and refine your prompts based on the model's responses.
  - Experiment with different phrasings to get the desired output.

5. **Critical Thinking:**

  - Evaluate the responses critically. Not all responses may be accurate or suitable.
  - Cross-check information when applicable, especially for critical or sensitive topics.

6. **Awareness of Model Limitations:**

  - Understand the limitations of the model. It may not have the most up-to-date information, and it might generate plausible-sounding but incorrect answers.

7. **Patience:**

  - Be patient, especially if the model provides incomplete or unclear responses.
  - Experiment with breaking down complex questions into simpler ones for better results.

8. **Ethical Considerations:**

  - Be mindful of the ethical use of the technology. Avoid generating harmful or inappropriate content.

9. **Technical Literacy:**

  - Understand the basics of how the model works to tailor your prompts effectively.
  - Be aware of potential biases in the model's responses.

10. **Adaptability:**

   - Be flexible in your approach. Adjust your questions or prompts based on the feedback from the model.

11. **Data Privacy Awareness:**

   - Avoid sharing sensitive personal information as the model might generate inappropriate responses.

12. **Feedback Utilization:**

   - Use the feedback provided by the model to guide subsequent prompts and improve the conversation.

13. **Learning Orientation:**

   - Continuously learn from the interactions with the model to enhance your skills over time.

14. **Experimentation:**

   - Explore different ways to structure your queries to understand the model's capabilities better.

Remember, while ChatGPT is a powerful tool, it's not perfect, and user engagement plays a crucial role in obtaining the desired outcomes. The more effectively you communicate with the model, the more useful and accurate the responses are likely to be.

Tricks

To get the most out of ChatGPT, users can employ certain tricks and strategies to enhance the quality of interactions and obtain more relevant and coherent responses. Here are some primary tricks:

1. **Experiment with Prompt Phrasing:**

  - Try rephrasing your prompts to see if you can get more accurate or detailed responses. Experimentation is key.

2. **Provide Context:**

  - Give sufficient context in your prompts, especially when asking complex or multi-part questions. This helps the model understand the specifics of your inquiry.

3. **Specify the Format:**

  - If you want information in a specific format (e.g., bullet points, a summary), explicitly mention it in your prompt.

4. **Use System Messages:**

  - Guide the model's behavior by using system-level instructions. For example, you can start your prompt with "You are an assistant that knows..." to set the context.

5. **Break Down Complex Queries:**

  - If your question is complex, consider breaking it down into simpler sub-questions. This can help the model provide more focused and accurate responses.

6. **Iterate and Refine:**

  - If the initial response is not what you're looking for, iterate and refine your prompts based on the model's output to guide it towards the desired information.

7. **Ask the Model to Think Step by Step:**

  - Request the model to think through a problem or provide information step by step. This can lead to more structured and detailed responses.

8. **Specify the Source of Information:**

  - Ask the model to generate responses as if it were a specific person or entity. For example, "Can you respond as if you were a historian?"

9. **Use Temperature and Max Tokens:**

  - Experiment with the model's temperature setting (higher values for more randomness, lower values for more determinism) and max tokens to control the length of responses.

10. **Leverage the Feedback System:**

   - Use the model's feedback feature to provide feedback on problematic outputs. This helps fine-tune the model and improves its performance over time.

11. **Handle Ambiguity Explicitly:**

   - If your question has ambiguous elements, explicitly specify your assumptions or ask the model to make certain assumptions before answering.

12. **Check for Consistency:**

   - If you're asking the model to generate a list or set of responses, check for consistency across the generated content.

13. **Use External Tools:**

   - If needed, use external tools or resources to fact-check or verify information provided by the model.

14. **Be Mindful of Sensitive Topics:**

   - Avoid asking the model to generate content that is inappropriate, offensive, or harmful.

Remember that while ChatGPT is a powerful language model, it has limitations, and user guidance is crucial for obtaining desired results. These tricks can help users navigate and make the most of their interactions with the model.

Prompts

The effectiveness of prompts in ChatGPT depends on the specific task or information you're seeking. However, here are some general prompt strategies that can be useful across various contexts:

1. **Open-Ended Questions:**

  - Start with an open-ended question to prompt a more detailed and informative response. For example, "Can you explain..." or "What are the key factors influencing..."

2. **Context Setting:**

  - Begin your prompt by providing context. For example, "In the context of [topic], can you provide information about..."

3. **Comparisons:**

  - Ask the model to compare or contrast different concepts, ideas, or approaches. For example, "Compare the advantages and disadvantages of..."

4. **Step-by-Step Thinking:**

  - Request the model to think through a problem or process step by step. For example, "Can you walk me through the process of..."

5. **Creative Writing:**

  - If you're looking for creative or imaginative responses, frame your prompt in a way that encourages storytelling or creative thinking.

6. **Ask for Pros and Cons:**

  - Solicit the pros and cons of a particular idea, solution, or approach. For example, "What are the advantages and disadvantages of..."

7. **Specify the Format:**

  - If you want information in a specific format, specify that in your prompt. For instance, "Provide a list of..." or "Summarize in three key points..."

8. **Role Play:**

  - Ask the model to respond as if it were a specific character, historical figure, or expert in a particular field. This can add a creative dimension to the responses.

9. **Socratic Questioning:**

  - Pose a series of questions to guide the model's thinking and encourage a more thoughtful response.

10. **Source-Based Queries:**

   - Ask the model to generate responses as if it were a specific source (e.g., a book, article, or person). For example, "Respond as if you were an expert on..."

11. **Clarification Requests:**

   - If the initial response is unclear, ask the model to clarify or provide more details on a specific point.

12. **Ask for Predictions:**

   - Request the model to make predictions about future trends, developments, or outcomes. For example, "What do you think will happen in the next decade regarding..."

13. **Explain Like I'm 5 (ELI5):**

   - Ask the model to explain complex topics in a simple and understandable manner, as if explaining it to a child.

14. **Challenge the Model:**

   - Encourage the model to think critically or consider alternative viewpoints. For example, "What might be the counterarguments to..."

15. **Use System Messages:**

   - Provide high-level instructions in system messages to guide the model's behavior. For example, "You are a helpful assistant with knowledge about..."

Remember to iterate on your prompts, refine them based on the model's responses, and experiment with different phrasings to achieve the desired outcome. The effectiveness of prompts often depends on the context and the specific information or task you're interested in.

Prompt vocabulary

When interacting with ChatGPT, you can use a variety of prompts and instructions to get the most useful and relevant responses. Here are some tips and useful words you can incorporate into your prompts:

1. **Specify the Format:**

  - "Provide a detailed explanation of..."
  - "List the steps to..."
  - "Summarize the key points of..."

2. **Ask for Clarification:**

  - "Can you elaborate on..."
  - "What do you mean by..."
  - "Please clarify..."

3. **Set the Tone:**

  - "Explain in a simple language..."
  - "Give a technical overview of..."
  - "Provide an example of..."

4. **Request Comparisons:**

  - "Compare and contrast..."
  - "Highlight the differences between..."
  - "Examine the similarities of..."

5. **Ask for Pros and Cons:**

  - "What are the advantages and disadvantages of..."
  - "Discuss the pros and cons of..."
  - "Evaluate the strengths and weaknesses of..."

6. **Define Parameters:**

  - "In the context of..."
  - "Considering the factors like..."
  - "With respect to..."

7. **Request Examples:**

  - "Can you give an example of..."
  - "Illustrate with a scenario where..."
  - "Provide a case study for..."

8. **Specify Time Frame:**

  - "Historically, how has..."
  - "In recent times, what changes have occurred in..."
  - "Looking into the future, what can we expect for..."

9. **Quantify or Qualify:**

  - "To what extent does..."
  - "How much impact does..."
  - "In what way does..."

10. **Ask for Opinions or Recommendations:**

   - "What is your opinion on..."
   - "Suggest strategies for..."
   - "Recommend best practices for..."

11. **Explore Causes and Effects:**

   - "What are the causes of..."
   - "How does [A] impact [B]..."
   - "Examine the effects of..."

12. **Discuss Implications:**

   - "What are the implications of..."
   - "Explore the consequences of..."
   - "Consider the impact on..."

13. **Request a Step-by-Step Guide:**

   - "Can you provide a step-by-step guide for..."
   - "Walk me through the process of..."
   - "Outline the necessary steps to..."

14. **Ask for Sources or Citations:**

   - "Are there any studies that support..."
   - "Can you provide references for..."
   - "Where can I find more information about..."

15. **Invite Creativity:**

   - "Imagine a scenario where..."
   - "What innovative solutions can you suggest for..."
   - "In a hypothetical situation, how would you address..."

Remember, the more specific and clear your prompts are, the more likely you are to receive helpful and relevant responses from ChatGPT.

Prompt generators

ChatGPT prompt generators are tools that help you create effective and engaging prompts for ChatGPT, a conversational AI model based on OpenAI’s GPT-4. ChatGPT prompt generators can help you customize your interaction with ChatGPT by setting a specific role, tone, style, and objective for the response.

Some of the best ChatGPT prompt generators that I found on the web are:

ChatGPT Prompt Generator | Welcome AI: This tool lets you experiment with different prompts for ChatGPT by providing some context and your goal or objective. It also follows OpenAI’s best practices for prompt design. ChatGPT Prompt Generator | Prompt Advance: This tool allows you to enforce the style of the response by choosing from a list of predefined roles, such as friend, teacher, coach, etc. It also lets you copy the prompt with one click. AI ChatGPT Prompt Generator | Taskade: This tool uses OpenAI’s GPT-4 model to generate engaging prompts based on your specified inputs. You can also adjust the temperature and top-k parameters to control the randomness and diversity of the response.

Prompt parameters

The term "prompt parameters" typically refers to the instructions or input provided to ChatGPT in order to generate a response. When interacting with ChatGPT, the input prompt is crucial in shaping the nature and content of the AI's output. The parameters can include the style of language, the level of detail required, the context for the response, and any specific instructions or constraints you want the model to follow.

Here's an example of how you might structure a prompt with parameters:

Without Specific Parameters:

   Prompt: "Tell me about climate change."

With Specific Parameters:

   Prompt: "Provide a concise overview of the causes and effects of climate change, focusing on the last decade. Include potential solutions and their implications."

In the second example, the parameters are more explicit, guiding the AI to deliver a response that meets specific criteria. The parameters help in obtaining more tailored and useful information from the AI.

It's important to note that while ChatGPT can understand and respond to a wide range of prompts, the clarity and specificity of your input can significantly influence the quality of the generated output. Experimenting with different parameters and refining your prompts can help you get the desired results.

Prompt techniques

Although I am familiar with Hugging Face's Transformers library, I don't have direct experience with HuggingChat. However, I can provide some suggestions for utilizing prompts effectively when working with similar conversational AI models within the Hugging Face ecosystem. These "prompt tricks" could help improve interactions with HuggingChat:

1. Contextual Prompting: Provide sufficient context before asking questions or giving input. This helps prime the model and ensures generated outputs consider previous information provided. For example: ```makefile <context>: Once upon a time, there were three little pigs who decided to build houses made of straw, sticks, and bricks respectively. A big bad wolf came... <query>: What did each pig do differently regarding building materials? ``` 2. Role Assignment: Define roles explicitly at the beginning of the interaction. This allows the model to understand its position and respond accordingly. Example: ```less System: You will now act as Shakespearean character Hamlet. User: Who art thou speaking to, dear friend? Assistant: To thine own self be true, good sir! ``` 3. Task Definition: Specify the task clearly so that the model understands how to react appropriately. Examples include Question Answering, Summarization, Sentiment Analysis, etc. Here's one for QA: ```markdown Question: Which American president appears on Mount Rushmore? Context: Four presidents—George Washington, Thomas Jefferson, Theodore Roosevelt, and Abraham Lincoln—are immortalized on Mount Rushmore. Answer: George Washington, Thomas Jefferson, Theodore Roosevelt, and Abraham Lincoln appear on Mount Rushmore. ``` 4. Scaffolding Technique: Break down complex queries into smaller sub-questions or steps. This approach makes it easier for the model to handle intricate requests. Example: ```python Query_Part1: Can you summarize chapter 5 of 'The Catcher in the Rye?' <summarization> Query_Part2: Now, please analyze Holden's emotional state described in the summary. <analysis> ``` 5. Error Handling & Repair: If unsatisfied with the response, rephrase the query instead of repeating verbatim. This encourages the model to look at the problem from another perspective. Example: ```vbnet Original Query: Please explain quantum physics briefly. Unsatisfactory Response: Quantum mechanics deals with atomic particles' behavior. Rephrased Query: Could you give me a concise overview of quantum theory concepts like superposition and entanglement? ``` These strategies should generally work well across different conversational AI models built on top of the Hugging Face Transformers library, including HuggingChat. Adapt them according to your particular needs to optimize interactions with the model.

Parameters

ChatGPT prompts can have various parameters that affect the quality and style of the generated responses. Some of the common parameters are:

length: This parameter controls the length of the response, measured in number of tokens or words. You can adjust this value to ensure the output is neither too short nor too lengthy. top_p: This parameter controls the diversity of the response, by filtering out low-probability words. Higher values lead to more diverse responses, while lower values generate more conservative and predictable responses. frequency_penalty: This parameter affects the likelihood of generating words that have already been used in the response. Higher values result in more varied responses, while lower values lead to more repetitive responses. presence_penalty: This parameter penalizes the likelihood of generating words that are not present in the input prompt. Higher values result in more relevant responses, while lower values lead to more irrelevant responses. stop_sequence: This parameter specifies a sequence of words that ChatGPT should avoid generating in its response. You can use this parameter to prevent the generation of inappropriate or sensitive content, or to maintain the focus of the conversation. You can learn more about these and other parameters from these sources:

Mastering ChatGPT Prompts: A Guide to Using Parameters by Bruce Lim How to Prompt ChatGPT - A Helpful Guide by allPrompts How To Write ChatGPT Prompts: Your 2024 Guide by Coursera 195 ChatGPT Prompts (& How to Write Your Own) by Semrush

Conclusion

Summarize the main points of the training, highlight the benefits and limitations of AI, and encourage the participants to explore more AI resources and opportunities.

Courses

(1) Artificial Intelligence Fundamentals Certificate | ISACA. https://www.isaca.org/credentialing/artificial-intelligence-fundamentals-certificate. (2) 18 Best Free AI Training Courses for 2023: Build Skills Now - Tech.co. https://tech.co/news/best-free-ai-training-courses. (3) Generative AI for Everyone | Coursera. https://www.coursera.org/learn/generative-ai-for-everyone. (4) Introduction to Artificial Intelligence (AI) | Coursera. https://www.coursera.org/learn/introduction-to-ai. (5) How to Learn Artificial Intelligence: A Beginner’s Guide. https://www.coursera.org/articles/how-to-learn-artificial-intelligence.

https://github.com/f/awesome-chatgpt-prompts https://promptbase.com/

Key terms

AI

AI system

Inference

Once trained, the model uses an inference algorithm to make predictions on new, unseen data. This involves applying the learned patterns and relationships to generate output based on input features.
The inference algorithms, particularly, are used to process data and perform tasks that #End-users prompt.

Intelligent machine

An intelligent machine refers to a device or system that possesses the ability to perform tasks that typically require human intelligence. This includes capabilities such as learning from experience, adapting to new situations, understanding natural language, solving complex problems, and making decisions.

Intelligent machines can take various forms, ranging from simple algorithms and rule-based systems to advanced artificial intelligence (AI) systems that utilize machine learning, deep learning, and other sophisticated techniques. These machines can be designed to perform specific tasks, such as image recognition, language translation, playing strategic games, or even driving vehicles.

The level of intelligence in a machine can vary, and it's often categorized into narrow or weak AI, which is designed for a specific task, and general or strong AI, which would have the ability to perform any intellectual task that a human being can. As of my last knowledge update in January 2022, we have achieved significant progress in narrow AI applications, but the development of general AI remains a complex and ongoing challenge.

ML

Training infrastructure

Efficient hardware and software infrastructure are necessary for training large-scale AI models. Without the appropriate infrastructure, training times can be prohibitively long, hindering the development process.

Deployment infrastructure

Once trained, #AI models need a robust deployment infrastructure to function in real-world applications. This includes considerations for scalability, reliability, and integration with existing systems.

See also