/ai-notes

notes for software engineers getting up to speed on new AI developments. Serves as datastore for lspace.swyx.io writing, and product brainstorming, but has cleaned up canonical references under the /Resources folder.

MIT LicenseMIT

AI Notes

notes on AI state of the art, with a focus on generative and large language models. These are the "raw materials" for the https://lspace.swyx.io/ newsletter.

This repo used to be called https://github.com/sw-yx/prompt-eng, but was renamed because Prompt Engineering is Overhyped.

This Readme is just the high level overview of the space; you should see the most updates in the OTHER markdown files in this repo:

  • IMAGE_GEN.md - the most developed file, with the heaviest emphasis notes on Stable Diffusion, and some on midjourney and dalle.
    • IMAGE_PROMPTS.md - a small swipe file of good image prompts
  • TEXT.md - text generation, mostly with GPT3
    • TEXT_CHAT.md - information on ChatGPT and competitors, as well as derivative products
    • TEXT_SEARCH.md - information on GPT3 enabled semantic search and other info
    • TEXT_PROMPTS.md - a small swipe file of good GPT3 prompts
  • INFRA.md - raw notes on AI Infrastructure, Hardware and Scaling
  • AUDIO.md - tracking audio/music/voice transcription + generation
  • stub notes - very small/lightweight proto pages
    • AGENTS.md - tracking "agentic AI"
    • CODE.md - codegen models, like Copilot
    • etc...
Table of Contents

Motivational Use Cases

Top AI Reads

The more advanced GPT3 reads have been split out to https://github.com/sw-yx/prompt-eng/blob/main/GPT.md

Beginner Reads

Intermediate Reads

Advanced Reads

Communities

People

This list will be out of date but will get you started. My live list of people to follow is at: https://twitter.com/i/lists/1585430245762441216

Quotes, Reality & Demotivation

Legal, Ethics, and Privacy

Alignment, Safety

  • Anthropic - https://arxiv.org/pdf/2112.00861.pdf
    • Helpful: attempt to do what is ask. concise, efficient. ask followups. redirect bad questions.
    • Honest: give accurate information, express uncertainty. don't imitate responses expected from an expert if it doesn't have the capabilities/knowledge
    • Harmless: not offensive/discriminatory. refuse to assist dangerous acts. recognize when providing sensitive/consequential advice
  • Just Eliezer entire body of work

Misc