Keep on to blur preview images; turn off to show them clearly
RT @OnlyXuanwo: The most interesting thing here is I didn’t add anything new, @ApacheOpenDAL already did this and I just combined them toge…
喜欢王小波，大概我们能成为朋友。 我的 2025 https://t.co/pAkSJnpKXA 我的 2024 https://t.co/HfDF6oduB7 我的 2023 https://t.co/QyV8PiZmOY ..............
RT @konstipaulus: Cursor for video editing doesn't exist for a reason. Since I've been tagged under this post multiple times and we have f…
Partner at a16z & AI enthusiast. Investor in @cursor_ai, @udiomusic, @replicate, @hedra_labs, @MistralAI, @character_ai, @tabulario, @_hex_tech, @labelbox, ...
I really hope my deep research system to re-identify peer reviewers was correctly submitted to ICLR. Still no news, weird.
Artisanal baker of reasoning models @pleiasfr
RT @SakanaAILabs: Sakana AI will be at #NeurIPS2025! 🐟🐟🐟 We’re pleased to announce that Sakana AI is co-hosting “AI for Science: Algorithm…
Building @SakanaAILabs 🧠
一图看懂 DeepSeek 新模型 DeepSeek-Math-V2！ DeepSeek 刚刚发布了 DeepSeek-Math-V2！ 没错，这是一个专门面向数学推理领域的领域专用模型，这次是基于 DeepSeek-V3.2 进行后训练/微调的, 意味着 DeepSeek 已经完全切换到新架构上了. 性能方面 IMO 2025 (国际数学奥林匹克竞赛) 刷到了 83.3%, 达到了 IMO 金牌水平. ProofBench-Basic 直接刷爆了, 得分99%, 碾压其他所有模型. 而 ProofBench-Advanced 仅次于 Gemini Deep Think (这个是带Agent的平台, 不是纯模型). 顺带, 今年IMO 2025第一也是我国. 然后我把 DeepSeek-Math-V2 这个 83.3% 的得分直接平均到每个问题, 得分应该是210分（满分252分，83.3%准确率），可以在美国（216分）和韩国（203分）之间，即第3名。
得分拟合到今年IMO榜单
Democratic politicians seem to be incapable of learning from Trump. They should never use the word "tariff". They should always say "federal sales tax". Trump went ballistic when Jeff Bezos suggested listing tariffs as if they were a sales tax. Using the word "tariff" is gross political incompetence.
deep learning