Needs:

Text to Text
- Google T5
- OpenAI GPT-*
- Nvidia Megatron-LM
- Microsoft Turning-NLG
- Bloom
  - https://bigscience.huggingface.co/blog/bloom
Text to Speech
- https://huggingface.co/models?pipeline_tag=text-to-speech&sort=downloads
- https://www.youtube.com/watch?v=aLBedWj-5CQ
Audio to Text
- Whisper
IRL Simulator
- image diffusion
- motion diffusion
  - https://anuragajay.github.io/decision-diffuser/
- motion prediction
- spatial mapping
  - https://twitter.com/LumaLabsAI
- dreambooth-style individual-based fine-tuning
  - identify participants in scenario
  - when simulating their replies, use this fine-tuned model
  - further fine tune with every interaction
  - figure out way to identify similar participants to known participants and make blended models as the starting point
  - maintain participant-indexed memory
- per-entity models: each encountered entity gets a dedicated model fine tuned on interactions with that entity
Judge
- Reinforcement-based learning
- The "cringe" circuit: continuously reasses past decisions and outcomes
- Training set
  - internet archive
  - project guttenberglakhglahglka
  - google books
  - movies
  - scentific renders
  - crafted simulations
  - https://twitter.com/du_yilun/status/1597618021342023680
- https://twitter.com/icodeblockchain/status/1599882951923466240

Misc Notes:

Provide feedback

Saved searches