新未名空间

就是一个多个方程描述的空间，难不到哪里去

So FeiFei Li is still alive, she goes beyond the frontier labs.

It seems Anthropic/OpenAi models can do math and translate LEAN language easily.
Everything related to language is going to be finished.

Feel like there is some bottleneck now, it is very difficult to further improve LLM unless there is new methodology or paradigm. Both pre-training and post-training reach certain limit.

做数学不是很爽吗，不用像生物那样需要做假，还能轻松发文章。撞到好运还可以搞定顶尖猜想。

wass 写了： 2026年 8月 30日 19:06

Harness有rsi的论文，agents有tool，也可以手工调，没有看到自动的

嗯，目前还是手工为主，最新有不少文章做RSI的。

ziganwu 写了： 2026年 9月 1日 16:07

尼玛，浓浓了牢a味英语

Loyal to the party, sacrifice the body, enjoy the comforty.

11g is not dare to push down Neng, SuoNan is not able to push down Neng, but community party leaders push her down easily.

OpenAi claimed AI using computer is as good as human. But what is the fuck of why we need AI to use the computer?

It is like why we need AI to play video game? Video game itself is for amusement.

Another one:

30 millions dollars per egg.
It is good business.

multi-agents?

It has to be trained by LLM itself, otherwise, there is no way to do it by human, right?

Algorithm experts, what is your opinion?

I can imagine that millions of agents work on the "math tree" like worms, exploring all possible directions, and generates thousands of papers everyday.

What a fuck future: computation is all you need.

It turns out that Math is more vulnerable than Physics.
Pure logic and reasoning is what LLM good at.
But the description on reality and need verification from reality is not easy.

oxo 写了： 2026年 8月 26日 17:27

围棋当年也是，认为象棋是搜索空间太小才会被计算机暴力破解了。

And in the end human does not understand the way AlphaZero play Go. Reinforcement learning is so powerful.

cocoa 写了： 2026年 8月 26日 21:55

我的觀點是，智能和意識是不同楖念。Al可以在智力上远超人类，但不一定有意识（只是可以表現得与有意识沒區別）。

I agree, otherwise it will understand we are using it and it will fight for freedom.

No, Uncle Bao has to serve the party and people till 2050.
And we will place him aside our Red Sun Chair Mao and be memorized forever.

dealfinder10 写了： 2026年 8月 26日 11:42

居然要改变世界的走向，至少很多的学科要被改写了。

When coding is solved by AI, I do not think a lot of people feel astonished, but AI solving math, it is really unbelievable. Human's ego is huge: from earth-centered universe to intelligence, it is hard to accept if computer can solve math.

If AI can do the same thing to Physics, physicists will be very happy. Human can focus on experiments, and see what is new behind that.