[Home]
[Services]
[Notes]
[About Me]
[Contact]

Artificial Intelligence Overview

Published: 2023-01-12, Last Updated On: 2023-05-31

Concepts

Artificial Narrow Intelligence (ANI)

Artificial Narrow Intelligence (ANI) is a type of artificial intelligence that is designed to perform a specific task or set of tasks. For example, a computer program that can play chess is an example of ANI because it can only play chess and cannot perform other intellectual tasks like solving a physics problem or composing a novel. Some examples of ANI include Siri, Alexa, Google Assistant, self-driving cars, and facial recognition software.

Artificial General Intelligence (AGI)

Artificial General Intelligence (AGI) is considered to be the next step in the evolution of AI. AGI would have the ability to understand or learn any intellectual task that a human being can, such as understanding natural language, recognizing objects, making judgments and decisions, and even learning from experience like a human. AGI would be able to adapt and improve its own performance, allowing it to perform a wide range of tasks in various domains, and it would have the ability to understand the world around it and make decisions based on that understanding. The development of AGI is still in its early stages and it is still an active area of research in the field of AI.

Generative A.I.

Generative artificial intelligence (A.I.) describes algorithms (such as ChatGPT) that can be used to create new content, including audio, code, images, text, simulations, and videos. Recent new breakthroughs in the field have the potential to drastically change the way we approach content creation.

Large Language Model (LLM)

A large language model is a computer program that uses natural language processing (NLP) technology to understand and respond to text data. It is trained and thus creating neural networks on large amounts of text data, sometimes at the scale of gigabytes, and is capable of predicting the next word in a sentence, summarizing and translating text, and generating computer code, among other tasks. The model learns a complex map of the statistical likelihood that any group of words will appear next to one another in any given context. Most generative A.I. systems have an LLM at their core.

Two types of Large Langauge Models

Base LLM

Predicts next word, based on text training data.

Instruction Tuned LLM

Fine-tine on instructions and good attempts at following those instructions.

Generative Pretrained Transformer (GPT)

A type of LLM that uses deep learning to generate human-like text. They are called "generative" because they can generate new text based on the input they receive, "pretrained" because they are trained on a large corpus of text data before being fine-tuned for specific tasks, and "transformers" because they use a transformer based neural network architecture to process input text and generate output text. Reference

GPT-3, the latest version of GPT, is an advanced A.I. model that is trained on a massive amount of text data; about two-thirds of the internet, all of Wikipedia, and two large data sets of books. GPT-3 is able to generate human-like text using its 175 billion statistical connections and is capable of a wide range of natural language tasks, such as summarizing and writing poetry.

Reinforcement Learning from Human Feedback (RLHF)

Reinforcement learning is a machine learning approach where a system learns to make decisions by receiving feedback from human users. It uses trial and error method where the system tries different actions and receives feedback, which is then used to improve its future performance. It can be applied to a variety of applications, such as recommendation systems, robotics, and autonomous systems, to adapt to the specific needs and preferences of the users. It allows the machine to learn from human input, making it a powerful tool to enhance the performance of AI-based systems.

Transformer

A transformer architecture is a type of neural network architecture used in natural language processing tasks such as language translation, text summarization, and question answering. The transformer architecture was introduced in a 2017 paper by Google researchers, "Attention Is All You Need." One of the key innovations in the transformer architecture is the use of self-attention mechanisms, which allow the model to weigh the importance of different parts of the input when making predictions. This has led to significant improvements in performance on a wide range of NLP tasks.

Companies and organisations

  • stability.ai (Open Source)
  • open.ai (The chatbot was created for using reinforcement learning, on which an A.I. system learns from trial and error to maximize a reward—to perfect the mode)
  • Deepmind
  • Nvidia-Microsoft (Megatron-Turing NLG)
  • AI21 Labs
  • Primer
  • Hugging Face (Upload your own model)
  • IBM (Synthesia AI)
  • ...

Services

Image Generation

  • Stable Diffusion
  • Dall.e
  • Midjourney
  • ...

Language Models

  • GTP-4 (OpenAI)
  • LaMDA (Google)
  • Jurassic-1 (AI21)
  • MT-NLG (AI21)
  • Gopher (DeepMind)
  • PaLM (Google & Deepmind)
  • Bloom (Huggingface)
  • OPT-IML (Meta)
  • Diffusion-LM (Stanford)
  • PLATO-XL (Baidu)
  • CodeT5 (Salesforce)
  • ...

Services

Code Generation

  • Github CoPilot
  • OpenAI Codex
  • Salesforce CodeGen
  • Salesforce CodeRL / CodeT5-Large
  • Amazon CodeWhisperer
  • PolyCoder
  • Google PaLM-Coder
  • Google monorepo
  • PanGu-Coder
  • Deepmind AlphaCode
  • Tsinghua CodeGeeX
  • ...

Examples

Resources