Only two ways that agents could really help me a lot:
A. We have a really deep talk step by step, then understand the every detail about something. The rest is just implementation.
B. We have a slightly deep talk like brainstorm, then I beg it could show some magic of real intellegence under my feedback loop harness I set for it.
Basically way A helped me more, yet I often slipped to way B because of my laziness.
10
If you hate somebody:
Spoil your daughter then marry him 🚫
Advise him buy the credits of claude 🫡
27
I still remembered when OpenClaw was very popular, crazy people in China were finding people to help them to install it, or finding tips and steps on social media.
Very rare people can build the OpenClaw completely under Claude Code.
Oxford researchers argue that LLMs can never invent anything.
It is mathematically impossible.
They published a paper called “Theory Is All You Need" and it argues against the claim that computational models can generate genuine novelty or new knowledge.
They analyzed the limits of generative ai, and the results are a brutal reality check for the idea that ai will replace human decision making under uncertainty.
Here is why AI is stuck and human cognition wins:
backward-looking vs forward-looking.. llms are probability machines that look backward at existing data. human cognition is forward-looking and capable of generating genuine novelty. human cognition operates theoretically "top-down" rather than "bottom-up" from data.
the "data-belief asymmetry".. the researchers use the invention of "heavier-than-air flight" to illustrate this concept. an ai relies on data-based prediction, which is largely imitative. humans, however, use theory-based causal logic that allows them to hold beliefs that go beyond existing data.
the intervention gap.. humans don't just process information; we use theory to practically "intervene" in the world. we engage in directed experimentation to generate entirely new data. ai-based models are theory-free and place primacy on existing data and prediction.
tldr?
AI uses a probability-based approach to knowledge and ia largely imitative. It can process data and make predictions, but human cognition relies on theory-based causal reasoning.
The decades-old analogy comparing human minds and computers to mere "input-output" devices is fundamentally flawed.
1
1
71
I tried many times to manage few agents at the same time, yet I found that my attention got burned once more than 3 agents.
9
I started to lose confidence to Claude Desktop because they still didnt fix the messy equation math expression issue on this application.
Codex app did this much better, which showed that the product team behind codex were really working and improving.
1
78
Claude Opus5 is the worst model that I ever used, it makes me hate reading its anwser.
The last time that I hate reading is everytime I read Karpathy's post because of his messy structure of expressing.
3
94
America, 2026
The parents cannot stop an American child from crying instantly.
Yet they simply whisper:
“China AI is comming”
The child immediately sleeps.
I do not support pacing frontier models. If I am going to die at the hands of killer AI, I want it to be American, not Chinese. Buy American, Die American.
43
claude fable 5给我的感觉是：我已经无法找到足够难的问题让fable发挥出超越opus的能力
过去这几个月自己打造的harness engineering几乎已经让整个体系形成有效闭环，从opus换成fable最大的体感只是响应速度变快了
过去几个月我愿意为了Claude本身的模型能力去付出超过其他模型N倍的溢价，但是放在现在的时间节点，DeepSeek V4 Pro的性价比真的有点超模了
1
97
苏炜杰老师也去 OpenAI 了
此前在交大有幸听过苏老师的报告，至今印象很深，受益匪浅。后来又读到一篇《苏炜杰：我从丘赛走来》，讲述他从农村一路走到北大数院，再到斯坦福读博的经历
这篇文章里最打动我的，并不是那些天才故事，而是他在斯坦福读博期间面临的压力：论文进展不顺，孩子出生带来的经济压力接踵而至。博士毕业时，银行卡存款几乎归零，还透支了一万五千美元信用卡。后来顺利拿到宾大的教职，才让这一切迎来了转机
基于自己的人生经历，苏老师对普通家庭出身的学生的建议非常值得一读：
“如果有志于学术，但是家庭经济一般，可能选择跟业界联系紧密、就业相对容易的领域是个明智的选择。”
对于计划进入业界的同学，他也提到：
“择业应该更多地考虑这个方向是不是在上升期，因为处于上升期的领域更看重个人能力和拼搏精神，对社会关系的要求较低。”
朴素、现实、智慧
历尽千帆，祝贺苏老师开始新的阶段
Personal update: I've joined OpenAI while on leave from Wharton. After a decade away, glad to be back in the Bay Area and train AI models here!
One more thing, I've been promoted to full professor, a decade-long journey made possible by many, especially my students.
1
185
you are too retarded to understand what Anthropic is doing.
they are training their own models by those Internet answer
27
Anthropic 又发了一篇关于中美 AI 竞争的文章。很多人又开始提 Dario 当年在百度工作的经历，推测他是不是因为在百度有过不好的体验，所以后来才对中国这么敌对
我觉得这个解释过于脸谱化了
更现实的解释是：Anthropic 如果想进入美国国家安全体系并希望持续发展在美国的toB业务，就必须完成身份切割。对一家美国 frontier AI 公司来说，对华强硬不是单纯的个人情绪，而是一种姿态。Dario 曾经在百度工作过，放在今天的中美 AI 竞争语境里，反而可能成为 Anthropic 必须反复处理的政治负资产。所以他们需要更强烈地证明：我们是站在美国国家利益这一边的
追根溯源，这件事让人想到魏则西事件
很多人把它简化成“百度害死了魏则西”。但如果认真拆责任链，里面至少有三层：涉事医院和莆田系医疗广告、医疗监管失灵、搜索平台的广告排序和审核责任
百度当然有责任，尤其是医疗广告、竞价排名、广告标识和风控责任。但把百度说成唯一主谋，反而遮蔽了更关键的医疗供给端和监管端问题
不是替百度洗地，而是判断一家公司，不能只看它在舆论里承受了多少情绪，而要看它在责任链条里到底处于什么位置
很多人对百度的看法，不是来自分析，而是来自一个早已固定的印象，这又催生了对 Anthropic 的错位看法
真正值得我们警惕的，不是立场不同，而是用私人恩怨解释结构性问题，用单一恶人解释系统性问题
这通常是认知偷懒
We've published a paper that explains our views on AI competition between the US and China.
The US and democratic allies hold the lead in frontier AI today. Read more on what it’ll take to keep that lead: anthropic.com/research/2028-…
1
67
Why doesn't democracy go for open source?
We've published a paper that explains our views on AI competition between the US and China.
The US and democratic allies hold the lead in frontier AI today. Read more on what it’ll take to keep that lead: anthropic.com/research/2028-…
1
23
最近重读吴翼老师的采访，读到这段突然想通了一件事：
纯 RL 训练（R1Zero）就像“梅西全场过人进球”——能做到，说明体系可行，但由于效率问题不会成为常态
对应到科研和交易里的 agents 系统也是一样：
当然可以把复杂流程拆成多个 agents，再设计 harness 进行运转，甚至让系统全自动化
但这样做的意义何在呢？
这实际上更像是一种 capability demo（秀肌肉）：证明体系能够自动化运转
真正落地时，harness 会遇到无数 edge cases，需要人工兜底、修补、干预
更现实的范式是：先把harness打磨到足够出色，再用 agents去放大增益，绝不追求全自动化
efficiency永远要在fancy前面
1
1
203