冯clarence

@wwwgoubuli 通过精神分析可以改变先验，可参考solm等人研究

我记得很多年前看到的一句话，人就是个贝叶斯系统，越往后越定型。

明天 volunteer 了在 Yang Zhang lab 组会上分享我自己的 AI loop。同一套 OS pattern 最近半年在 automath 和 newmath 跑出 3,400+ 0-axiom Lean 4 定理、5×/周自动 release，多篇文章在 peer review，还有合作项目在同步推进，也接到了我 PhD 的 TCR / pMHC 上（JSI-Pi，joint scaffold-interface posterior）。 我们 lab 也是个老牌 top-tier research lab —— I-TASSER 在 CASP automated server 赛道长期 top，170k+ 注册用户、160 个国家。但据我观察，AI 工具的使用密度还远没有到它能到的位置 —— 主流用法还停在把 ChatGPT 当成 search / 助力工具，真正能放大产出的 agent loop、多模型协作、自动化 pipeline 几乎没怎么铺开。这也是我想做这次分享的原因。 slides 做都做了，干脆 share 出来 —— 看看大家能不能从里面收获点什么，也欢迎大家和我交流讨论，多多合作。 里面总结了几条经验： · 怎么搭一个减少幻觉的自动化 loop：能结构化的全部结构化，能 gate 的全部 gate，AI 只在 bounded task 里跑。很多幻觉不是模型突然变差，而是任务边界、输入 schema、输出 format 和 review gate 没有收紧。 · 怎么用多机器 + GitHub 做隔离协作：private repo 也可以，每改一次都 commit，随时回滚，所有人的贡献都可视化；多台机器常开，各自跑各自的 loop，最后通过 branch / PR / merge 汇合。 · 怎么蒸馏自己的"经验库"：受 gstack 的 office-hour 启发，我把一位资深研究者公开发表的论文、talk、quote 爬下来，加上自己长期和很多人交流的笔记，一起丢给 AI 蒸馏成一份"私人 PhD office-hour"。它帮我判方向、补盲点 —— 当我在想一个模型时，它会提示我"导师大概会问的下一个问题"、"我现在没考虑到的角度"。这是一种超过自身经验的经验。 我相信科研品味和直觉，在某种程度上是可以被 AI 捕获、被 scale 的。 我做这次分享有两个目的：一是希望大家能更多用上现在的 top AI；二是希望实验室能合作一起产出。个人和单个 agent 的力量都有上限，只有协作起来并行推进，效率才会加倍 —— 每个人都能参与，每个人的参与都在 GitHub 上可视化，方向由我们把控，剩下的让 AI 迭代。未来一定会发生一次范式转移（或许已经发生），做 AI 的 lab 更应该走在时代前面，去善于使用工具。协作跑起来，每个人的产出都会更快、更好，包括我自己。 完整 slides：researchloop.lexaverse.dev

@Stanleysobest 对自己感觉的怀疑是悲剧的开始

我感觉文化不高的人说什么话开头都是—— 我感觉！

The more I study psychoanalysis, the more I feel like this field is a collection of empirical summaries. At the same time, they have developed a theoretical framework atop these observations that is as complex and convoluted as the Ptolemaic system. truly difficult to explain.

In the era of the Brain-Computer Interface, psychotherapists are still talking about concepts like integration, growth, wholeness, and maturity. Doesn't that seem particularly absurd?

I've recently realized that in society, many people are telling you their story. If you don't have a story of your own, they'll try to weave you into theirs. Pay attention to the most popular sci-fi narrative of the moment...

kind of feel recently people just get crazy about the dev of ai, but actually the most benefit thing to do is enhance the original everyday workflow, instead of getting fomo and rushing to the ai industry

I think people somehow get crazy about the development of ai and I think the um most benefits things people can do is actually um enhancethe original everyday workflow with ai instead of a rush into this ai industry umunder the pressure of ai fomo

ai+心理学赛道目前有科研团队招ra吗？我是心理咨询师，长程来访者，同时会python vibecoding 能手搓简陋agent，持续学习中，找一个靠谱的科研团队

I believe that partial replacement will occur within the next two years, while complete replacement will likely happen once AI systems with a world model emerge.

As a psychotherapist who has also undergone more than 150 hours of psychotherapy, I am now fully convinced that psychotherapists will be replaced by AI.

@jayvanbavel this is good

You are being manipulated: RageCheck is a free tool that analyzes online content for manipulative framings—language designed to provoke emotional reactions rather than inform. Social platforms reward engagement, and outrage generates more engagement than nuance. This creates incentives for creators to frame information in emotionally provocative ways, regardless of whether that framing is accurate or fair. RageCheck helps you see these patterns so you can make more informed decisions about what to believe, share, and engage with. See if elites are using manipulative framings: ragecheck.com/about

Want to find something interesting here about psychoanalysis and neuroscience and brain science