Keep on to blur preview images; turn off to show them clearly
can we appreciate the insanity of “Older hardware underperformed in evals and was removed” for what is conceptually a lot of parallel General Matrix Multiplications
achieve ambition with intentionality, intensity, & integrity - @dxtipshq - @sveltesociety - @aidotengineer - @latentspacepod - @cognition + @smol_ai
Safe spaces online, for humans.
Research Scientist @meta (FAIR), Prof. @Unige_en, co-founder @nc_shape. I like reality.
是的，AI 编程还是 Mac 好使。学员遇到环境安装的基本 90% 都是 Windows。 零基础 AI 编程的朋友，非常推荐去苹果店取完自己 Mac 后，搭配一套我的课程！好工具➕好课程，能少走很多弯路🌝
我的AI编程课(https://t.co/HVZn3ItASW) |B站up主 | 分享创造 + 无限迭代ing
or do we already do that because we are minimising KL with more recency bias rather than over the entire distribution?
RL and efficient distributed pretraining • eXperiments lab • memes and training lores
*monotone* oh no, not the briar patch!
We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
why do we accept approximating distribution from data/model with forward KL (left picture)? why not work towards algos that look like (right picture)?
or do we already do that because we are minimising KL with more recency bias rather than over the entire distribution?