cacio-e-pepe

5 ms·

1. 1.▲
   by cacio-e-pepe 11d ago

   Could you expand? Curious.
2. 2.▲
   by cacio-e-pepe 11d ago

   relevant: https://mathstodon.xyz/@tao/117219101339291693
3. 3.▲
   by cacio-e-pepe 2mo ago

   Nice, looking forward to the report. And thanks, huge pasta fan :)
4. 4.▲
   by cacio-e-pepe 2mo ago

   > So we did mechanistic studies on small models, Gemma 4 particularly, and found the hidden state for different layers carry meaningful self-awareness signal for various situations. Neat! Just to make sure I understand - you trained your
5. 5.▲
   by cacio-e-pepe 2mo ago

   Honestly, stellar performance by the model at the capability being measured.
6. 6.▲
   by cacio-e-pepe 2mo ago

   Okay so the diffusion model generates image assets, and you iterate on those, and then another model turns your favorites into code, is that right?
7. 7.▲
   by cacio-e-pepe 2mo ago

   Looks interesting. I'm a bit too busy to read the docs, but I'm curious - how does it work?
8. 8.▲
   by cacio-e-pepe 2mo ago

   What are the core knobs for a RAG pipeline? See any interesting patterns? Eg in practice, what did the agents tend to tune? Did domain matter?