Random ideas & research

Theme examples

• https://twitter.com/danielvlopes/status/1785777942049772011

Prompt library examples

• https://prompts.ray.so/raycast
• https://www.yohana.com/#things-we-do
• https://www.rabbit.tech/promptbook.pdf
• Improving tricky prompts: https://www.anthropic.com/news/prompt-generator

Tools for verifying facts (help with reasoning)

• https://github.com/Libr-AI/OpenFactVerification
• https://github.com/nilsherzig/LLocalSearch

There is an open source privacy project called: https://docs.searxng.org/ It apparently takes your query and searches a bunch of other search engines real time, and mixes the results. I still can’t easily find out how it works since it’s clearly not using the API for each of these search engines. Maybe they keep an up-to-date page structure?

The spirit of this project is “don’t use publicly hosted instances of searxng, host your own” but for fun I found a public one so you can try it: https://baresearch.org/

And this LlocalSearch project is using this prebuilt docker image of SearXNG: https://github.com/searxng/searxng-docker so by adding this to their docker compose their project gets a self-hosted API for search! And it exposes a whole bunch of APIs which match other APIs, here’s the google-styled APIs it exposes: https://docs.searxng.org/dev/engines/online/google.html

Similar projects

• ChatBotUI
• https://recurse.chat/
• https://news.ycombinator.com/item?id=39532367
• https://www.reddit.com/user/Zer0Tokens
• https://msty.app/
• https://docs.openwebui.com/
• https://bento.me/jordi

Promotional lists

• https://opensourcealts.com/project/chatgpt
• https://github.com/billmei/every-chatgpt-gui
• https://saner.ai/90-places-to-launch-your-ai-saas-startup-continuously-update/

Agents

• Minecraft agent: https://github.com/MineDojo/Voyager
• There were multiple examples of scapers using that chrome plugin that adds keyboard shortcuts to every button, find that
• Agent to write articles: https://t.co/5OQDcaFwdN
• sweep.dev — automating junior developer
• open source like sweep: https://twitter.com/swyx/status/1657892220492738560?lang=en
• https://paperswithcode.com/sota/code-generation-on-humaneval
• https://huggingface.co/spaces/bigcode/bigcode-models-leaderboard
• SWE Agent: https://swe-agent.com/
• OpenDevin: https://github.com/OpenDevin/OpenDevin
• https://plandex.ai/
• https://www.openinterpreter.com/
• Automating workflows: https://www.tasker.ai/

There are now many open source projects that support writing a project from scratch, such as:

• https://github.com/Pythagora-io/gpt-pilot2
• https://github.com/geekan/MetaGPT3
• https://github.com/AntonOsika/gpt-engineer4
• https://github.com/smol-ai/developer
• How can ChatGPT be effectively utilized for reading library source code, resolving coding issues, and serving as a dedicated coding assistant tailored for a specific repo?

Browser control:

• Skyvern: https://bookface.ycombinator.com/posts/79706
• https://app.multion.ai/playground

Tools

• Search for AI: https://tavily.com/
• Reader version of any page: https://simonwillison.net/2024/Jun/16/jina-ai-reader/
• Code sandbox: https://e2b.dev/

Starter projects

• https://jumpstartrails.com/#features
• Bullet train

Custom instruction ideas for different personalities / agents

• https://news.ycombinator.com/item?id=40474716&utm_source=hackernewsletter&utm_medium=email&utm_term=ask_hn

Voice

• First voice computer: https://www.youtube.com/watch?v=1ZXugicgn6U

• LLM computer: https://x.com/karpathy/status/1723140519554105733

• O1 project: https://github.com/OpenInterpreter/01

• https://retellai.com

• In browser WASM whisper (a little slow but no API): https://github.com/ggerganov/whisper.cpp

• A faster re-implementation of Whisper. This is the core of many other projects: https://github.com/SYSTRAN/faster-whisper

  • There are speaker separation libraries built off this
  • There is also an OpenAI Whisper compatible API server Docker image built off this
  • Collabra: agent built on Faster Whisper: https://github.com/collabora
  • Whisper real-time: https://github.com/davabase/whisper_real_time

• Ultravox (direct audio LLM): https://simonwillison.net/2024/Jun/10/ultravox/

• Picovoice (hardware running): https://picovoice.ai/blog/chatgpt-ai-virtual-assistant-in-python/

• WebRTC: https://github.com/domchristie/webrtc-hotwire-rails

• Notable HN discussion about Jarvis: https://news.ycombinator.com/item?id=38682095

• Amazing integrated demo: https://www.soundhound.com/voice-ai-products/platform/

• Training my own model (especially for kids, Ismail knows them): https://www.speak.com/blog/asr-levelup

• Lots of coverage of different voice models: https://blog.unrealspeech.com/

Running LLM in ruby/browser

• In-browser Whisper: https://whisper-turbo.com/
  • https://www.ratchet.sh/whisper-turbo
  • https://x.com/fleetwood___/status/1701567799942934644
• In-browser LLM: https://summize.fleetwood.dev/
  • https://playground.fleetwood.dev/
• Echo cancellation
  • https://webrtc.github.io/samples/src/content/getusermedia/record/
  • https://github.com/varuncm/echo-cancel
• Wakeword
  • https://github.com/Picovoice/porcupine
  • https://github.com/OzymandiasTheGreat/wakeword-zero
  • https://github.com/dscripka/openWakeWord

Local LLM

• 30gb of ram for Gemma-27b: https://huggingface.co/bartowski/gemma-2-27b-it-GGUF
• 9b: https://huggingface.co/bartowski/gemma-2-9b-it-GGUF

Business tools

• Support emails: https://helptail.com/

Component library

Material design Tailwind

• https://github.com/rinturaj/tailwind-material-3
• https://github.com/aribudin/tailmater
• https://www.material-tailwind.com/docs/html/button
• https://tailwindcomponents.com/component/tailwind-css-button-ripple-effect-by-material-tailwind

Material design components NOT tailwind

• https://github.com/material-components < official google library
• https://material.angular.io < official google fork for angular
• https://www.mdui.org < popular other library
• Reference: https://m3.material.io/components/cards/specs

Novel LLM interfaces

• Topic explorer: https://globe.engineer/
• Game tool (sent to David)
• LLM for circuit design (did call with the guy)