For years now, every developer is how to put up with hearing about how AI is going to make them obsolete and steal the they’ve been putting on their family.
多年来,每个开发者都在忍受关于 AI 将如何让他们过时并偷走他们赖以养家糊口的东西的言论。
But besides the occasional identity crisis, I feel like most of us are actually doing pretty well.
但除了偶尔的身份危机,我觉得我们大多数人实际上过得还不错。
Sadly, are. Mathematician friends may not be so lucky, because over the last three weeks, AI has killed more open math problems in the entire human race, managed in the previous decades.
遗憾的是,我们的数学家朋友可能就没那么幸运了,因为在过去三周里,AI 攻克的开放数学问题比整个人类在过去几十年里解决的还要多。
One 80 year old conjecture from a list of the hardest problems of the 20 first century got the promotion during the sanctity.
一个来自 21 世纪最难问题清单的 80 年历史猜想在这段时间内被推翻。
And another 30 year old graphic conjecture fell to the programs.
另一个 30 年历史的图论猜想也被程序攻克。
And 58 problems and things have gotten so bad.
还有 58 个问题,事情变得如此糟糕。
Santa tow, a man who won an international math Olympiad medal at age ten, while the rest of us were making cinema sugar sandwiches and watching rocket power, it stood up at the International Congress of mathematicians and warned of a crisis in the foundations of mathematical values and practices.
陶哲轩(Santa tow),一个十岁就赢得国际数学奥林匹克奖牌的人,当我们其他人还在做肉桂糖三明治、看《火箭动力》时,他在国际数学家大会上站起来,警告数学价值观和实践的基础出现了危机。
In to this video will look back at the summer mathematics fl two machines and find out what that may mean for the rest of us.
在本视频中,我们将回顾这个夏天数学如何败给机器,并探讨这对我们其他人可能意味着什么。
It is the August and 2026, and you are watching the code report.
现在是 2026 年 8 月,你正在观看《代码报告》。
And this whole AI doing math things started about a year ago once the model sold five of the six problems at the international math limpidly.
而这一切 AI 做数学的事情大约始于一年前,当时一个模型在国际数学奥林匹克(limpidly)中解决了六道题中的五道。
But at the time, it felt mostly harmless, because competition math is the Lecoq of math problems, and at the time and the the the model to prove the the, the, the, the, the, the the dish conjecture about unit distances and suddenly it didn’t feel so Hmmm anymore.
但当时,这感觉基本无害,因为竞赛数学就像是数学题中的 LeetCode(Lecoq),而当时模型被要求证明关于单位距离的 Erdős 不同距离猜想(dish conjecture),突然间它就不再让人觉得那么“嗯”了。
Couldn’t believe I had trouble sleeping for the dish conjecture.
难以置信,我因为 dish 猜想而失眠。
And June 16, researchers from 15 universities published what they call the light detector, which politely asked the world to establish some God rails and AI and mathematics research, which the, the, the mostly ignored.
6 月 16 日,来自 15 所大学的研究人员发表了他们所谓的“光探测器”,礼貌地请求世界为 AI 和数学研究建立一些护栏,但世界大多忽略了。
And last month, the Americans were pretending like soccer, and Europeans were pretending to the medieval infrastructure without air conditioning.
上个月,美国人假装喜欢足球,欧洲人假装住在没有空调的中世纪基础设施里。
Mathematician event apogee, opposed to a counter example to ask that disproved the Jacobean conjecture, which is one of the most famous problems in the era of geometry.
数学家 event apogee 提出了一个反例,推翻了雅可比猜想(Jacobean conjecture),这是几何领域最著名的问题之一。
The idea is that if you have a polynomial function whose derivable checks out at every single point, meaning it’s perfectly reversible anywhere you zoom in, then the entire function should be reversible too.
其思想是,如果你有一个多项式函数,其导数在每个点都成立,意味着无论你在哪里放大它都完全可逆,那么整个函数也应该是可逆的。
Now, I know that probably doesn’t mean much to you, but since 1933, nine, that that elt so obviously true of mathematics that it made smells list of the hardest problems of the for a century, because nobody could actually prove or disprove it, that the until event did last month with some help from favor.
我知道这可能对你来说意义不大,但自 1939 年以来,数学家们觉得它显然是真的,以至于它被列入斯梅尔(Smale)的 21 世纪最难问题清单,因为没人能真正证明或推翻它,直到上个月 event 在朋友的帮助下做到了。
And just two days later, it happened again with a 30 year old graph theory and called the dense Garg golds conjecture.
仅仅两天后,又有一个 30 年历史的图论猜想——稠密 Garg golds 猜想——被攻克。
But what you can think of, like a delivery routing problem, if shipments are allowed to split across multiple rounds, the question is whether you can always bundle them into single unspun rounds without the total delivery cost going up.
你可以把它想象成一个配送路线问题:如果货物允许拆分到多条路线,问题是是否总能将它们合并成单条路线而不增加总配送成本。
And for 30 years, everyone assumed that you could, they just can improve it.
30 年来,每个人都假设可以,只是无法证明。
In a research named the memory, Robin pointed gp t 5.6 added and basically told it to make a breakthrough.
一位名叫 Robin 的研究人员让 GPT-5.6 去解决它,基本上就是让它取得突破。
The model responded with a tiny network of the seven note nine edges that are forced. Every shipment into single route was always more expensive than letting them them split.
模型回应了一个由七个节点和九条边组成的微小网络,该网络迫使每批货物走单一路线总是比拆分更昂贵。
It is proved in the conjecture, and it’s not just the prose doing this.
它推翻了这个猜想,而且不只是专业人士在做这件事。
That same week, a Columba P, H, D student knocked out six open edis problems in five days, and the 23 year old amateur banned out another.
同一周,一名哥伦比亚大学博士生在五天内解决了六个开放的 Erdős 问题,一名 23 岁的业余爱好者又解决了一个。
And this is mostly before any of the big tech labs took their own shots.
而这大多是在任何大型科技实验室自己出手之前。
But then just two weeks ago, open A, I got in on the ball, in which they removed that internally with the next major mode, solve the open problems across math and Theoretical Computer Science.
但就在两周前,OpenAI 也加入了,他们透露内部用下一个主要模型解决了数学和理论计算机科学中的开放问题。
And they even shipped all them to get up, and they y’ve wen certificate, which is a formal proof the compiler can check ck mechanical.
他们甚至把所有成果都发布到了 GitHub,并且还获得了证书,这是一种编译器可以机械检查的形式化证明。
And the most famous one on the list is what I know since have thought about a live, which is how tightly you can crm identical size balls into a space.
名单上最著名的一个是我一直思考的球体填充问题,即你能把相同大小的球塞进空间有多紧密。
And the general bound on this problem hasn’t moved since 1078. But open A, I just improved it.
这个问题的通用界限自 1978 年以来一直没有进展。但 OpenAI 刚刚改进了它。
The other nine problems include things like the first explicit construction of a non soft group, and they disprove of cons regidor conjecture, which I won’t pretend to understand.
其他九个问题包括首次明确构造非 sofic 群,以及推翻 cons regidor 猜想,这些我不会假装理解。
The mathematicians have been chasing some of them sense the nineties.
数学家们自 90 年代以来就一直在追寻其中一些问题。
But of course, Dorio couldn’t just sit by and get big derivative by Sam is.
但当然,Dorio 不能坐视不管,被 Sam 抢了风头。
So just last week, anthropic fire back by announcing that one of its own unreleased models made progress on the women hypotheses, which is one of the most famous unsolved problems, and all of mathematics.
所以就在上周,Anthropic 反击,宣布其一个未发布的模型在黎曼猜想(women hypotheses)上取得了进展,这是所有数学中最著名的未解决问题之一。
It is 100 hundred and 67 year old question about the distribution of prime numbers, and it one of the seven millennium price problems with a literal million dollar bounty on it.
这是一个关于素数分布的 167 年历史问题,也是七个千禧年大奖难题之一,悬赏高达一百万美元。
And the best part is that it wasn’t an anthropic mapei tion who discovered it. It was geared summoner, the creator of bung js.
最棒的是,发现它的不是 Anthropic 的员工,而是 geared summoner,Bun.js 的创建者。
It is on his job for his unfiltered rich now and need to get fit.
他正在做他的工作,因为他的 unfiltered rich now 和需要健身。
Two problems during his run. Jar red, ask club to take a real stab approving it so it did, and generated 650 wrong ideas for solving the problem.
在他的运行过程中有两个问题。Jar red 让 club 真正尝试证明它,于是它做了,并生成了 650 个错误的解题思路。
Jared then told it to try again. And this time the model spent a day a half, the ordinary and 60 soap agents inside cloud code run in 20 400 shell commands, writing hundreds of Python scripts and burn in 31 million of peoples.
Jared 然后让它再试一次。这一次模型花了一天半时间,在 cloud code 中协调了 60 个 soap 代理,运行了 20,400 条 shell 命令,编写了数百个 Python 脚本,消耗了 3100 万个 token。
Now that still didn’t lead to approving the hypothesis, it it bump the frax solutions that provably satisfied the hypotheses from 41% to 67%, which I’m told this pretty ground breaking.
这仍然没有导致证明该假设,但它将可证明满足该假设的解的比例从 41% 提高到了 67%,我听说这是相当突破性的。
The result was validated by two mathematicians at employ C. A, checked by two outside number theory experts, and formalized and lean.
该结果由 employ C. A 的两位数学家验证,由两位外部数论专家检查,并在 Lean 中形式化。
We live in crazy times, but stuff like this wouldn’t be possible without amazing cloud providers.
我们生活在疯狂的时代,但如果没有出色的云服务提供商,这样的事情是不可能的。
Railway, the sponsor, today’s video, they once again refused to waste your time with a full add, and you can say thank you by checking them out for your next project at the link below.
Railway,今天视频的赞助商,他们再次拒绝用完整广告浪费你的时间,你可以通过下方链接查看他们用于你的下一个项目来表达感谢。
And this has been the code report. Thanks for watching, and I will see you in the next one.
以上就是本期《代码报告》。感谢观看,我们下期再见。