aichannel

we were discussing how inefficiencies of english inflate tokens and maybe a minimalist language form is needed and highly powerful

I’m so excited to announce Gemma 3n is here! 🎉

🔊Multimodal (text/audio/image/video) understanding
🤯Runs with as little as 2GB of RAM
🏆First model under 10B with @lmarena_ai score of 1300+

Available now on @huggingface, @kaggle, llama.cpp, https://t.co/CNDy479EEv, and more https://t.co/Xap0ymhCmr

The race for LLM "cognitive core" - a few billion param model that maximally sacrifices encyclopedic knowledge for capability. It lives always-on and by default on every computer as the kernel of LLM personal computing.
Its features are slowly crystalizing:

- Natively multimodal

Andrej Karpathy

we were discussing how inefficiencies of english inflate tokens and maybe a minimalist language form is needed and highly powerful 
https://x.com/karpathy/status/1938626382248149433

i dabble with ai using open identity data from crypto https://alexpaden.tech

prev startups & racecars

@ network school

planting causal loops; harvesting reflexivity

did you see the new hierarchical paper and the Mixture of Recursion from deepmind?

i think both of these is what they did to win the math medals

adaptive compute

harder tokens more compute

another topic of discussion was ideas as tokens

i was exploring a similar idea with lemmas and a DSL/programming language


main point is it searches theorems so shorter tokens = more compute to explore