Top 8 A.I. Terms Defined

How to Become a "Dinner Table A.I. Expert”

1. Introduction to AI

• AI: AI, or artificial intelligence, is the development of computer systems that can perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.

• AGI: AGI, or Artificial General Intelligence, refers to a type of artificial intelligence that is as smart as a human across the board. This means it can understand, learn, adapt, and implement knowledge in a wide range of tasks, just like a human can. It's the concept of a machine that could perform any intellectual task that a human being can do.

• Prompt: A prompt is the input given to the model to generate a response. It serves as a cue or a trigger for the model. For example, when interacting with a language model like GPT-3, the prompt might be a question or a statement that you want the model to respond to or elaborate on. The model's response is based on the patterns it has learned during its training process.

2. Machine Learning and Deep Learning (ML & DL)

• Machine Learning (ML): A type of AI that enables a system to automatically learn and improve from experience without being explicitly programmed. ML algorithms use statistical methods to improve their predictions or decisions over time.

• Deep Learning (DL): A subfield of ML that uses artificial neural networks with multiple layers (hence 'deep') to model and understand complex patterns in data. DL models are especially good at handling large, high-dimensional datasets.
  

3. Natural Language Processing (NLP)

• NLP: A field at the intersection of computer science, AI, and linguistics that focuses on the interaction between computers and human language. It involves making computers understand, interpret, and generate human language in a valuable way.

4. Large Language Models (LLMs)

• Language Models: Language models are types of AI models that are trained to understand, generate, and manipulate human language. They are often used in tasks like translation, summarization, and question-answering.

• Large Language Models (LLMs): LLMs are a type of language model that has been trained on an extensive amount of data and has a large capacity to understand and generate text. GPT-3 by OpenAI is an example of an LLM.

5. Training and Fine-Tuning Models

• Training Models: This is the process of feeding a model data and allowing it to adjust its internal parameters to learn patterns in the data. 

• Fine-tuning Models: Once a model is trained, it can be fine-tuned on a specific task or dataset. This involves continuing the training process on a more narrow set of data or a specific task to make the model more effective in that area.

6. Popular Applications of Language Models

• OpenAI: This is the most popular Large Language Models (LLMs) available on the market today for both consumer and developer use. OpenAI has two main sides to their products/ services. Consumers and Developers. As a consumer we can use applications like ChatGPT and DALL-E which are applications built by OpenAI on top of OpenAI’s API. As developers we can use OpenAI’s API to create our own applications and send/ receive data from OpenAI without having to build out our own LLM. 

• ChatGPT: This is a chatbot developed by OpenAI that uses their LLM to understand and generate human-like text. This is the “consumer-level” version of OpenAI.

• Google Bard: This is a chatbot developed by Google that uses their internal LLM and data from Google search to generate up-to-date, human-like responses.

7. Technical Terms Related to A.I.

• Structured Data: This term refers to data that is organized in a predefined manner or according to a specific model. It is usually stored in databases with clear, identifiable patterns that make it easy to sort, process, and analyze. Examples of structured data include numbers, dates, and groups of words and numbers called strings (like names or addresses), which are typically stored in relational databases or spreadsheets.

• Unstructured Data: In contrast, unstructured data is data that doesn't fit into predefined models or isn't organized in a pre-defined manner. It's typically text-heavy, but can also include data like images, videos, email messages, social media posts, and more. While it's more difficult to process and analyze due to its lack of structure, unstructured data can contain valuable insights that structured data might miss.

• Vector Database: These are specialized databases designed to efficiently store and query high-dimensional vector data. Vector data, in the context of machine learning and AI, often represents features or embeddings extracted from raw data. This could be anything from the numerical representation of a word or sentence (in natural language processing), to the features of an image (in computer vision). Vector databases are optimized for performing operations like nearest neighbor search, which is crucial for many machine learning tasks, where you want to find the data points in your database that are most similar to a given input.

• Embeddings: Text embeddings, also known as word embeddings, are a type of representation for text where words or phrases from the vocabulary are mapped to vectors of real numbers. These vectors capture much of the semantic information of the words, such that words that are similar in meaning will have similar vectors (i.e., their vectors will be close in the embedding space). This makes text embeddings very useful for natural language processing tasks, as they allow models to work with text in a numerical and computationally efficient form.

• Transformers: Transformers is the backbone of how AI models are really good at understanding and working with language. The special thing about Transformers is that they can look at a whole sentence or piece of text at once, instead of just one word at a time. This means they can understand how words relate to each other in a sentence, which makes them great at tasks like translating languages or answering questions. They do this by giving each word a score (or weight) based on its importance in the sentence, helping the model to figure out what the sentence is really about. For example, in the sentence "The cat, which is black, sat on the mat", a Transformer would understand that "cat" and "sat" are related, even though there are other words in between.

• API (Application Programming Interface): An API is a set of rules and protocols for building and interacting with software applications. It's a way for different software systems to communicate with each other. It defines the methods and data formats that a program or service can use to interact with the software component, much like how a user interface enables a person to interact with a software.

• Endpoint: An endpoint in the world of APIs is a specific address (usually a URL) where a particular function or service can be accessed. It's like a specific phone number that you call to reach a particular person or service. For example, a weather API might have one endpoint (URL) for getting today's weather and another endpoint for getting the forecast for the next week. When you or a program wants to get that information, you "call" that endpoint by sending a request to its URL.

• POST: In the world of web APIs, POST is a request method supported by HTTP used to send data to a server to create a new resource. It's like writing a letter (data) and sending it (POSTing it) to a specific address (the endpoint). The server at that address receives the letter and uses the information in it to create something new, like a new user account or a new database entry.

• GET: GET, on the other hand, is a request method used to ask the server for a specific resource, like asking for a page of a website or the current weather from a weather API. Using the letter analogy, it's like sending a request to a library (the server) asking for a specific book (the resource). The library finds the book and sends it back to you.

• Webhook: A webhook is a way for an app to provide other applications with real-time information. A webhook delivers data to other applications as it happens, meaning you get data immediately as opposed to a typical API where you would need to poll for data frequently in order to get it real-time.

• AI Hallucinations: This term is often misused when referring to AI responses. In reality, AI models don't perceive or imagine things as humans do, so they can't truly hallucinate. When an AI, like a large language model (LLM), gives an output that seems made up or unrelated to the input, it's not because the AI is hallucinating, but because it's generating responses based on patterns it learned during training.

• AI Confabulation: This term is a more accurate description of the phenomenon where an AI produces an output that is plausible but incorrect or unverifiable. This can occur when the model has incomplete information or when it tries to generalize from the patterns it learned during training to a new situation. The AI isn't intentionally making things up; it's just trying to provide the best response based on its programming and the data it was trained on.

8. Technical A.I. Applications

• LangChain: Langchain is an open source framework that builds on top of the capabilities of large language models (LLMs) like ChatGPT by introducing prompt templates, vector databases, autonomous agents, and external tools to build more sophisticated and autonomous AI applications. It serves as the backbone of applications like AutoGPT and BabyAGI, allowing chaining of multiple prompts, memory retention for context, and the use of autonomous agents to interact with external tools. 

• AutoGPT: One of the first applications built using Langchain that allows users to create A.I. bots that can perform tasks automatically and then use the results of that task to influence and create the next task. It uses the concept of agents to build an autonomous bot to go from A-Z without having to constantly send prompts. The downside with AutoGPT is it’s still very much a research/ developer tool. There is not a friendly user interface like ChatGPT. 

• Stammer.ai: Stammer.ai is a white-label AI software platform that enables agencies to create and resell custom, AI-powered apps like chatbots and lead generation tools.

• CoPilot: Copilot is a tool that helps developers write code by suggesting lines of code based on their current context. It is powered by OpenAI's GPT-3 language model, and it can be used with a variety of programming languages, including Python, Java, C++, and JavaScript.

• Hugging Face: Hugging Face is a popular choice for NLP tasks because it offers a wide range of pre-trained models that can be used for a variety of purposes. Its most popular product is the Hugging Face Transformers library, which contains pre-trained NLP models that can be used for a variety of tasks, such as text classification, sentiment analysis, and question answering. 

• Redis: Redis is an open-source, in-memory data store that is often used for caching and data retrieval. It is a popular choice for technical AI applications because it is fast, scalable, and easy to use.

• Replit: Replit is a cloud-based development environment that allows users to create, share, and collaborate on code projects. It is a popular choice for technical AI applications because it is free, easy to use, and has a large community of users.