Twitter Beğeni Hilesi

Seeing convergent evolution between Arctic Snowflake Hybrid MoE vs DeepSeekMoE. Both enable overlapping dense computation with MoE chip-to-chip communication overhead. Both can learn common knowledge on a shared path. And, unlike many other MoE examples, both use 100++ experts.

In talking to policy makers and AI researchers, I realised there's a fact agreed upon by all researchers, but understood by almost no policy makers. This uncomfortable fact is why AI policy is hard.

Had a fun time helping host the Agents 2.0 Hackathon this past weekend! Presented some new tools from MultiOn with Pranav Putta