As someone who's not well-versed in AI, how much of it is actually truly transfo

As someone who's not well-versed in AI, how much of it is actually truly transformative and how much of it is hot air? I've got some friends in mathematics who are seriously depressed after the Jacobian conjecture got disproven because their jobs might be at serious risk.

作为对AI不太懂的人，有多少是真变革，多少是吹牛？我有几个数学系的朋友，雅可比猜想被证伪后心情沉重，因为他们的饭碗可能岌岌可危。

On the other hand, i also hear people getting fired and replaced by "AI" which is worthless and just leaving the remaining employees with more work.

另一方面，我也听到有人被辞退，被没用的“AI”顶替，结果留下的员工活儿更重了。

What's your take on it? Where do you see AI in 5-10 years?

你怎么看？你觉得AI在未来5到10年会走向何方？

图片: https://i.4cdn.org/g/1786017174465794.jpg

can't predict 5 or 10 years because ai moves so fast. quite a lot of jobs are probably at risk. which? not sure.

没法预测5年或10年，因为AI发展太快了。不少工作可能危险。哪些？不好说。

附件: https://i.4cdn.org/g/1786019288480025.mp4

>>109477938

Holy shit they put an allosaurus in space?

卧槽，他们把异特龙送上天了？

>>109477945

with ai yes. and it can obviously be done better.

有AI啊，而且显然能做得更好。

>>109477756

I can only give my perspective as a software developer and I think it will (or already has) completely changed how people work.

我只能以软件开发者的角度说，我觉得它已经（或即将）彻底改变人们的工作方式。

I buy into the narrative that it is another layer of abstraction, as compilers have been in the past.

我认同那种说法，它是另一层抽象，就像过去的编译器一样。

Regarding 5-10 years prediction, I think our jobs will revolve mostly about harness engineering, loop engineering or whatever you wanna call it. Basically writing scripts to run LLM cronjobs.

说到5到10年的预测，我觉得我们的工作主要会围绕“驾驭工程”、“循环工程”或随便你怎么叫。基本就是写脚本跑LLM定时任务。

>>109477756

It will absolutely mean the number of openings for mathematicians declined very rapidly. The students are right to think they won’t get jobs. This plus funding cuts means it is over for most of them. The ones who do get jobs will unfortunately be forced to compete by producing huge amounts of work with LLM assistance.

那绝对意味着数学家的就业机会急剧减少。学生们担心找不到工作是对的。加上经费削减，对大多数人来说这行算完了。能获得职位的人，不幸也得靠LLM辅助铺天盖地产出，才能拼个你死我活。

>t. mathematician, unfortunately

>本人数学家，不幸得很

>>109477756

Sounds a lot like someone's homework essay assignment...

听起来挺像某人作业论文的题目……

I prefer not to answer this shitpost.

这玩意儿我不答，纯水帖。

>>109478024

Mathematics were never a mainstream market. You don't know shit either.

数学从来不是主流市场。你也不懂个屁。

Maybe in the US you can be a "mathemacian" if you aren't sub 90 IQ, but elsewhere in the world it's a very niche area for gifted people. Statistics != mathematics.

可能在美国，智商不低能当“数学家”，但世界其他地方，这是天赋异禀者的冷门领域。统计学不等于数学。

>>109478051

I recommend at least spelling a word you’re putting in scare quotes correctly if you want to tell me about my own job, and maybe waiting until you’ve learned to read too.

想对我这行指手画脚，我建议你至少把引号里的词拼对了，顺便学会读字再说。

>>109478071

quant is still hiring

量化交易还在招人。

>>109477756

Likely the biggest global recession of all time.

很可能是史上最大的全球衰退。

However, they'll probably introduce UBI (universal basic income), so you'll be allowed to vegetate in your 5x5 feet cage while struggling to get water, because another AI datacentre will be built next to your block. Billionaires and trillionaires will own everything. Nothing on the internet will be real anymore.

不过他们多半会推行全民基本收入（UBI），让你在自己那5x5英尺的小格子里混混日子，还得为水发愁，因为另一座AI数据中心就建在你住的那片儿旁边。亿万富翁和万亿富翁会拥有一切。互联网上将再无真实可言。

Several jobs are safe, for now, but they'll be quickly diluted, because mathematicians, teachers, etc. will quickly re-train themselves into plumbers, street sweepers, etc. With the oversaturated market, it'll be difficult for anyone to get a job.

有些工作暂时还算安全，但很快就会被冲淡，因为数学家、教师这些人会迅速转行去做水管工、扫大街之类的活儿。市场一饱和，谁想找份工作都难。

Future is so bright I don't need my eyes to see it.

未来太光明了，我都不用眼睛去看。

图片: https://i.4cdn.org/g/1786023816548193.png

>>109478051

You've got such a small penis that it's unbelievable, and I'm not even someone who likes maths or mathematicians.

你那玩意儿小得难以置信，而且我本来就不是喜欢数学或数学家那类人。

AI is going to Make Mathematics Great Again!!!

AI要让数学再次伟大！！！

Once they build the first T800, it's all over.

等他们造出第一台T800，一切就全完了。

图片: https://i.4cdn.org/g/1786024116448754.jpg

>>109478250

Maybe, but from what I hear everyone is saying all the usual non academic jobs are also getting extremely competitive. When I referred to funding cuts I meant the actual academic system, rather than some other mathematical job. The lack of funding plus AI is likely to kill careers for a lot of people. I think it’s unlikely LLMs don’t eventually have a similar level of impact on at least some other jobs, although this might be an especially good environment for them.

也许吧，但我听说现在所有那些非学术的普通工作也都卷得不行。我提到经费削减指的是真正的学术体制，而不是别的什么数学岗。经费短缺加AI很可能会扼杀很多人的职业生涯。我觉得LLM最终不对至少一部分其他工作产生类似冲击的可能性不大，尽管这环境对它们来说也许特别有利。

>>109477756

>depressed after the Jacobian conjecture got disproven

>雅可比猜想被证伪后抑郁了

This was Anthropic's marketing response to OpenAI's original marketing strategy of presenting "proofs" for some mathematical problems. Even on the Wikipedia page for the Jacobian conjecture, it notes:

这是Anthropic对OpenAI最初那种给数学问题摆“证明”的营销策略的回应。连维基百科上雅可比猜想的条目都写着：

>The correctness of the counterexample is easy to verify with any computer algebra system. It has not been revealed, however, how it was found. Nevertheless, it led some mathematicians to elaborate on the mathematical reasons and the implications of the existence of the counterexample.

>反例的正确性用任何计算机代数系统都很容易验证。不过，它是怎么被发现的并未公开。尽管如此，它还是让一些数学家深入探讨了该反例存在的数学缘由及其启示。

Levent Alpoge is also a skilled mathematician. Even if his counterexample found with Fable turns out to be valid, it's more demonstrative of the fact that the operator of an LLM needs to know what the fuck they're talking about to do anything productive with it.

Levent Alpoge也是一位技艺精湛的数学家。就算他和Fable一起找到的那个反例最终被证实有效，这更多地说明了一点：操作LLM的人得知道自己在扯什么玩意儿，才能用它干出点正经事来。

I've been a professional programmer for almost 10 years now and my biggest concern is people new to programming are increasingly retarded. They don't know what the fuck a STACK TRACE is, they just copy/paste shit into Claude and hope it spits something out that works.

我当专业程序员快十年了，最担心的就是新入行的越来越废。他们根本不知道堆栈跟踪是个什么鬼，光会往Claude里复制粘贴东西，盼着它吐出一段能用的代码来。

I use LLMs at work and stay up to date on new model releases, and like to try running different models locally. The more you use LLMs the more you realize what exactly they're useful for, and the hype surrounding them (largely from people who don't use them) just becomes fucking insane to witness.

我在工作中用LLM，也关注新模型发布，还喜欢在本地跑各种模型试试。LLM用得越多，你就越明白它们到底适合干嘛，而那些围绕它们的炒作（多半来自根本不怎么用的人）简直离谱到让人看着都发疯。

Just yesterday Fable 5 failed to vertically align 2 fucking buttons for me. LLMs as a programming tool are here to stay and can definitely be useful but I very much fear a future competency crisis for programmers.

就在昨天，Fable 5连对齐两个破按钮都没给我搞定。LLM作为编程工具肯定会留下来，也确实有点用，但我非常担心未来程序员的胜任力危机会爆。

图片: https://i.4cdn.org/g/1786024835182559.png

>>109478393

off-topic but does anyone else think the T800 is the coolest fictional machine ever made?

跑题了，但有没有人觉得T800是有史以来最酷的虚构机器？

图片: https://i.4cdn.org/g/1786025822782562.jpg

>>109477756

Nobody knows for sure. You will sometimes see overconfident CS people who think they know everything just because some parts of the transformer rely on matrix multiplication and the (first) training objective is to predict text, but turns out these things actually imply very little. I wouldn't place too much weight on the predictions of those who are well-versed in AI.

没人能确定。你有时会看到那些过于自信的计算机科学家，觉得自己无所不知，就因为Transformer某些部分靠矩阵乘法，而且（最初的）训练目标是预测文本，但这些其实说明不了什么。我不太信那些AI行家的预测。

I'm a mathematics master's student, and I'm part worried, part not. It's true that AI is starting to deliver some impressive results which I wouldn't be able to do myself, but mathematics isn't just that. It remains to be seen whether it can conjecture/intuit and introduce new concepts rather than just proving or disproving theorems.

我是个数学硕士生，一半担心一半不担心。确实，AI开始出些我自个儿做不到的亮眼成果，但数学不止这些。它能不能猜想/直觉感知并引入新概念，而不只是证明或推翻定理，这还得看。

I think mathematics, like software engineering, is "AGI complete", in the sense that to really solve it end-to-end, you need to develop a human level AI, at which point all white collar work would be over anyways. AI can reduce the demand for certain jobs without being human level by making the top performers more efficient, but still I think the effect would be similar everywhere.

我觉得数学，跟软件工程一样，是“AGI完备”的，意思是想从头到尾真正搞定它，你得开发出人类级AI，那时所有白领工作反正也完了。AI不用达到人类水平，光靠提高顶尖人才效率就能减少某些岗位需求，但我觉得影响哪儿都差不多。

If I had to guess, in 5-10 years we'll get there.

要我猜的话，5到10年内就能到那步。

>>109478565

Humanoid robot? Yes. Machine? No. That'd be an Eva unit or giant robot, portal gun etc

仿人机器人？行。机器？不行。那得是EVA机体或巨型机器人、传送枪之类的。

>>109478830

Hi,

> If I had to guess, in 5-10 years we'll get there.

>要我猜的话，5到10年内就能到那步。

I ask to be sure I understand you. Do you mean to say that in 5-10 years, we'll have human-level AGI that matches or exceeds the absolute best world-class human experts in mathematics/software engineering?

我问是想确认我没理解错。你意思是，5到10年内，我们会有达到或超过世界级数学/软件工程顶尖专家水平的人类级AGI？

Kind regards,

此致敬礼，

>>109478830

>borderline toddler who learned to stand on his own 2 days ago scared about job prospects in math after learning of the calculator

>一个刚学会站两天的小屁孩，听说有计算器了，就怕数学饭碗不保

seems legit

看着挺靠谱

Why are mathfags always the single most profoundly retarded demographic? Aren't mathtards supposed to be smart?

为啥数学宅总是最蠢的一个群体？数学呆子不该是聪明的吗？

>>109479855

>Do you mean to say that in 5-10 years, we'll have human-level AGI that matches or exceeds the absolute best world-class human experts in mathematics/software engineering

>你意思是，5到10年内，我们会有达到或超过世界级数学/软件工程顶尖专家水平的人类级AGI

Yes. I do not have any real arguments though, it's just my opinion.

对。但我没啥硬论据，就是个人看法。

>>109479974

I'm not scared, read my post again.

我没怕，你再读读我帖子。

>>109481899

Forgot to mention, I am not very confident in my prediction because we probably need at least one more breakthrough to reach AGI. LLMs have a memory problem. They can remember things you said within their context window, but forget completely as soon as that context window runs out unless you update their weights. Updating the weights is an expensive, slow and difficult process unlike whatever happens in our brains. Things like LoRA exist for this reason, but these techniques are closer to temporary hacks than a real solution.

忘了说，我对这预测不太有信心，因为要达到AGI可能还得有至少一个突破。LLM有记忆问题。它们能记住你上下文窗口里说的，但窗口一过就全忘，除非更新权重。更新权重又贵又慢又难，跟咱大脑里发生的不一样。LoRA这类东西就是为这存在的，但这些技术更像临时补丁，不是真解法。

I doubt that the core learning algorithms will be optimized to the point where they are as fast as our brains soon. But maybe we can still get better memory through some smart context window fuckery, like storing a compressed non text-based representations of the history in the context window.

我怀疑核心学习算法短期内很难优化到和大脑一样快的程度。但也许我们还能靠点聪明的上下文窗口操作来提升记忆，比如在上下文窗口里存压缩后的非文本历史表征。