开启时会模糊预览图，关闭后正常显示
2021到2025年，代码大型语言模型（Code-LLMs）及相关生态系统的演变概览。
AI驱动的代码生成中，编程开发和研究领域的演变。
Incidentally, I only now noticed that DeepSeek's «Generative Reward Model (GRM) empowered by Self-Principled Critique Tuning (SPCT)» checkpoints were released at some point. This still used Gemmas and V2-lite + V2.5 as teacher. Imagine how good their GRMs are when based on V3.2.
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
RT @VoidAsuka: Claude buying Bun. It hit me today: the AI era is compressing the entire developer-tooling stack into the hands of a few pla…
ai agents @hud_evals | owned @AIHubCentral (1 million users,acq.) ex climate protester🦦 dont do the deferred life plan
GRM was a big deal. Few paid attention, even though it was close to DeepSeek's post-R1 hype wave. Math-V2, V3.2 use its ideas. This will be slowly acknowledged. Innovations are remarkably slow to diffuse.
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
Give a man his dream job and you will sleep well with dreams. Running YC was @garrytan’s dream. Happy for my brother who is clearly killing it.
Founder: @mixpanel Pizzatarian, programmer, music maker
This is basically a list of labs without the Mandate, fast-followers. > RLVR gives small open-source models an edge Always overviews of the last year's alpha. GRM not cited. Math-V2 doesn't seem "very novel". I think it'll matter as much as Math-V1, in retrospect.
GRM was a big deal. Few paid attention, even though it was close to DeepSeek's post-R1 hype wave. Math-V2, V3.2 use its ideas. This will be slowly acknowledged. Innovations are remarkably slow to diffuse.