Keep on to blur preview images; turn off to show them clearly
RT @_Yuming_Li: 30% off all my painting materials on Gumroad until the end of October — use code FALL30 My painting course stays 32% off 💫…
Founder/CEO @Gumroad
what do the cool kids use for images of code snippets today? I use carbon dot sh but I've seen nicer ones around
building workers observability @cloudflaredev, prev founder @baselimehq (acquired by cloudflare), prev aerodynamicist
Actually no: we read a ton of Chinese papers Chinese people very kindly write them in English, with occasional (but these days, rare) mistakes They don't really *have to* bother anymore, you know. It's just that Arxiv etc is convenient But we don't read Zhihu it can get worse
Well, I overstate for rhetorics, obviously they have the “ulterior” motive of being cited and building personal scientific reputation outside of China, and even in China citations in foreign papers are valued. English has become the lingua franca But it doesn't cover 100%
性能评估显示：gpt-oss-safeguard 在同时执行多个政策下，准确率优于 GPT-5-thinking 与其他 gpt-oss 模型。 详细内容：
学AI找小互，找小互，上 https://t.co/4PVaHEr5r3 ...
OpenAI 发布开源安全推理模型：gpt-oss-safeguard 专门用来进行内容审查 模型版本： gpt-oss-safeguard-120B gpt-oss-safeguard-20B 不同于传统“黑盒”内容审核算法： gpt-oss-safeguard 以推理代替记忆，能直接读取你的的安全政策，并在推理时遵循该政策执行判断进行内容审查。 无需学习... 这代表着内容安全从“被动学习规则” 进入“主动理解规则”的新阶段 意味着： - 政策不需要再重新训练，只需修改输入文本即可。 - 模型能输出自己的推理过程，说明为什么作出判断。 - 不同产品、不同场景可以自定义不同政策。新。 - 不同产品、不同场景可以自定义不同政策。
性能评估显示：gpt-oss-safeguard 在同时执行多个政策下，准确率优于 GPT-5-thinking 与其他 gpt-oss 模型。 详细内容：
if Reddit is still the goldmine of data, we'll need faster scrapers
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