hintymad

I recall several mathematicians (possibly including Terence Tao) mentioning that fields in mathematics have become so specialized and isolated that a conference like the ICM feels more like a collection of mini-conferences. An expert in one area can barely understand a talk in another.

Modern AI feels like a godsend to mathematicians. It helps them break down boundaries and connect concepts in ways a mere mortal couldn't imagine.

The examples in the posts suggest that the past does not contain all the patterns, or information in general, about the future. If so, isn't it natural that point forecast will fail in some cases?

So, so, so much politics and public reputation management. Zhang should have never had to suffer like that.

Very true. Unfortunately, when there are people, there will be politics. I remember when reading Yau's autobiography, I kept marvel how much calculation, or "politics" if you will, that Yau mentioned or implied in the book.

My guess is the outcome would have been the same for the same reasons?

At least Zhang didn't have to spend 7 years working on the Jacobian conjecture. He said in an interview that he always wanted to work on number theory. Moh asked him to work on Jacobian, and he obliged.

The Jacobian Conjecture

Interestingly, Yitang Zhang of the twin-prime-conjecture fame spent 7 years working on the Jacobian conjecture under the advisor Tzuong-Tsieng Moh at Purdue. A key step in his thesis used a corollary of Moh's. It turned out that the corollary was incorrect. As a result, Moh refused to write any recommendation letter for Zhang, and Zhang couldn't find any teaching or research job and ended up spending years working at a Subway[1].

Imagine Zhag had ChatGPT in 1986 when he started working on the Jacobian Conjecture.

[1] Of course now this has become an inspiring story. That said, the story definitely invokes complex emotions. The best way to describe it is probably this Chinese poem, which I have no idea how to translate: 庾信平生最萧瑟，暮年诗赋动江关

- PC office productivity software destroyed expensive professional products.

I agree with the lesson too. Just to be precise, wouldn't the current model war be more akin to open-source office suite versus MS office suite? If so, then the cheaper option didn't really win. That said, the open-source alternatives didn't really feel the same as MS Office, and it took them a long time to reach the feature parity (or did they ever?). In contrast, the open-weights models are getting close enough to the SOTA models, and users can easily switch from one to another without feeling any difference for mojority of the tasks.

I wonder how Chinese companies can make their models so much cheaper than the US companies. I'm not sure government subsidies are the answer. Subsidizing a single company with a few billion dollars, maybe. Subsidizing at least three companies with 10s of billions of dollars annually? Do we have proof of that? I assume we can't pin it on the lower cost of engineers in China, either. The top engineers are not that cheaper, and isn't engineering cost a small fraction of the cost of the model companies? Besides, if engineering cost is the driving force, can we really say that the US companies have a technical edge?

open models is what will kill Anthropic and OpenAI.

Maybe killing Anthropic is a good thing? Anthropic believe that they are the moral god and they get to determine what we can do and can't do, and they get to tell us what model to use and what not to use. I find it very counterproductive.

If we look at how AI solves those Erdos problems or IMO problems, we can see a clear pattern: AI is like a perfect cramming human being: it has seen and memorized pretty much all the known problem sets and the solution patterns. It knows all the areas of maths - a feat that even the best mathematicians in the world can achieve. So AI can link an obscure solved problem in one area to a hard problem in another, leading to amazing new solutions[1].

I think we can use AI similarly: asking AI for ideas that we haven't thought of before, and asking AI to connect the dots in new ways. To do these two effectively, though, we will need deep conceptual understanding of what we work on, and strong intuition to further refine the answers given by the AI.

That is, we don't offload our thinking. We just augment it with AI. It's like playing the game of ABC: given a letter, and name all the movies starting with it. One can be very familiar the movies but still fail to recall most of the titles. AI can exactly help with that type of recall.

[1] That's why OpenAI's model could solve the hardest problem (problem 6?) in 2025's IOI, but failed to solve the easiest one for human (problem 3?).

There is a difference between reading a short article and reading an actual book. Reading short articles gives you quick facts and situational awareness, but it doesn't build wisdom.

My personal theory is that wisdom comes from deep, long-form reasoning. You can't get it just by skimming short articles or memorizing isolated facts. When we read a deep and comprehensive book, the real value is the mental labor. It forces us to follow a long, complex argument, spot subtle patterns, and actively debate the author in our heads. We have to weave different facts together to see the bigger picture.

This kind of "hard reading" actually changes our brain by building new neural connections. Wisdom goes beyond collecting information. It's more about the cognitive journey of thinking deeply and widely about a topic. Short articles give us data points, but books train us how to think.

Of course, serious researches may tell us otherwise. I don't have any data points outside my personal experience.

But, at token rates, 10x or 100x the cost of open models or what I was spending on the frontier models a month ago

And we can't ignore the power of "good enough". GLM5.2 may not be as good as the SOTA models, but it can be good enough for most, of not all, of our needs.

Looks very useful. Thanks!

Yeah, large volume of comprehensible input is the key. It's just that the frequency of words is of Zipfian distribution, so I thought maybe flash cards could give me more frequent exposure of the important words in less time.

For language learning, I wish there was an audio-first flashcard app that changes up the example sentences every time. Right now, I'm using Anki to learn Japanese vocabulary from N5 to N3[1]. I know the words and the example sentences well enough to read N3-level text, provided I know the grammar. But when it comes to listening, I struggle to understand even N4-level spoken Japanese. Anki just doesn't offer enough variety for me to truly internalize what the sound means in different contexts. Plus, seeing the text before hearing the audio tricks my brain. I think I'm learning the sound, but it's an illusion because I already know the meaning from seeing the word first.

[1] I feel like Anki offers diminishing returns once you get past N3. Advanced words usually have subtle nuances that you can only really pick up through rich context, like in a full paragraph or a TV scene. Native-speaking kids can understand complex words in context because they have a deep grasp of a smaller, simpler vocabulary. That’s why I’m focusing on mastering high-frequency, simple words first to build a learning flywheel. I'm hoping this will eventually let me pick up new words naturally through reading and listening, just like a native kid does.

You're right, there are nuances in different policies. I was referring to the general power and consent that Europeans grant to the EU council. In my naive view their power is unchecked. As a result, we can start with good intent and good regulations, but eventually they will abuse their power as its the nature of power.

Honest question: when Europeans give so much power to EU and usually favor regulations by the government, isn't it natural that the government will try to implement more control? And it looks EU officials do not have to accountable for anything. They will not suffer personally even when their policies wreck havoc. I don't quite understand why Europeans can trust EU at all. Case in point, EU HQs shut down its air conditioning on floors 1 through 7 to prevent electrical overloads, leaving the upper levels used by top officials unaffected. Yet did anyone like Leyen get punished? Note I'm not naive enough to believe politicians don't have special treatment in other countries. But at least in some countries, politicians will not be so shameless that they'd do it in broad day light.

It's great that we have yet another competing models. The more models we have, the less likely we are subject to the ideologies and the controls thereof by the cults like Anthropic. And of course, it drives down the cost of tokens.

penalties for white collar criminals is zero tolerance

I think this is more about punishing political grafting than white-collar crime.

Wouldn't toxics like nitrites accumulate over the years? Also, I'd assume the purpose of perpetual soup is to concentrate the aroma and the taste, but is there going to be a diminishing return?

Remember it was reported that OpenAI didn't think that ChatGPT would be successful? OpenAI thought that ChatGPT was yet another toy before its launch. Yet once ChatGPT became an overnight success, Altman started to talk about how AI would be dangerous, how it would displace or even replace jobs. In contrast, Amodei seemed to always believe in what he said. So, can we say that Altman is a opportunistic businessman, and Amodei is a cult leader?

I'm surprised that people still take Gebru seriously. She is a disgrace to the community because she always, I mean literally always, attacks her critics by motives. You think bias is a data problem? You're a bigot (See her dispute with LeCun). You disagree with my assessment on an ML model? You are white male oppressor (her attacking a Google's SVP). Oh, did I mention that she even said that some loss functions are more racists than others on X?

Gebru is not a researcher. She is a modern-age Trofim Lysenko, who politicizes everything and wields political correctness as a weapon to purge any dissent.

This seems a pretty distinct corporate culture in the US. The companies pamper you in the boom, but dump you in the bust. When times are good, they shower you with perks yet when times are slightly bad, they lay you off like a replaceable line item. This is quite different from Europe or Japan, for instance.

Nice. It's great to see democracy work. Now do the California officials everywhere. Oh wait, my bad. Constituents are happy with Karen Bass and the like. Never mind.

It's wild that Europeans either accept this kind of Soviet-style nonsense, or just feel like they can't do anything about it. Even citizens of the USSR saw right through that blatant, Orwellian 'some animals are more equal' hypocrisy, yet Europeans are somehow just... okay with it? People like Ursula von der Leyen say all the right things while driving the continent into stagnation and crisis, yet they never have to face any personal consequences. It's truly amazing.

That means tech companies won't hire engineers in Michigan if they are required to be oncall, as many such engineers rotate their oncall schedules?

I largely blame people like Amodei for such outcome. As product owners, they could've done it the old way: telling people how great this product is, how much potential it has, and what kind of guardrails the companies are building and etc. But oh no, Amodei has to do the doom trolling 24x7, while in the meantime plays a cult leader by telling people only he knows how to the guarding angel of the AI or the humanity thereof. Ironically, the same people also push their companies to develop more powerful AI in full speed. They think ordinary Americans are so stupid that they can't see through them?

I'm now being paid to go to school, and get raises every year until I'm fully ticketed (way more than I ever made in the entry tech positions).

This. People tend to underestimate the joy that steady progress brings. A quick peak usually just leads to a long, depressing decline. Many people would rather take a career that grows a few percent every year for four decades over one that spikes and crashes any day. It’s better to be a slow grower who stays valuable than a flash-in-the-pan who burns out by 35[1].

[1] Honestly, I think there is a reason for this decline that has nothing to do with AI: the IT industry has just matured. Aside from the classic GoF patterns and Enterprise patterns and their variations, what new popular and deep design patterns have we actually adopted lately? Or look at all those must-know data structures and algorithms that are all over the web. How many of those were invented in the last ten years? Even in open source, where are the new platform-level projects invented in the last 5 years that every major company is pouring resources into? There are not many.

In other words, we are just eating our own tail at this point. It is just CRUD to death. When things get this stagnant, tech departments inevitably turn into cost centers. Even without AI, we were already heading toward a dead end. AI just happens to be the tool that makes it easy to automate everything because, at the end of the day, most of our work is just rearranging the same old code patterns anyway.

I'd rather see a coding agent that can follow steps in a plan file to a T while following guardrails and adhering to the proper coding conventions in the human reviewed spec.

In fact, I'd rather see Anthropic publish a convincing project that does this using Claude. The project should be complex enough and novel enough to show the world how reliable and powerful Claude is. That is, Anthropic does not need Amodei or its employees to tell us that whatever percent of engineers will lose their jobs. They can just show us. Easily.

I assume that "you didn't warn us" and "why are you so pessimistic" would lead to different scale of anger.

There are a few possible explanations of Dario and Anthropic's behavior, if you are not sure if all they want is pump up the valuation like me:

- Anthropic has a cult-like culture. AI Safety is their religion. The AI Constitution is their bible. Dario is their cult leader. Employees are the apostles. They just really really believe their church, and only Dario is qualified to manage the AI safety.

- Asymmetrical risk. If Dario speaks optimistically about AI and he turns out to be wrong, he'd face the rage of many people. If he fans the doomerism of AI and he turns out to be wrong, at most he will be mocked.

- Regulatory capture. After all, pretty much all the AI big shots in Biden government went to Anthropic. They produced the Biden's regulation, and they made it clear that they wanted to pick a few winners to back.

This is hardly surprising. Think about Staff+ engineers: their work is a lot like commanding AI. Most climb the corporate ladder through hard work, excellent engineering skills, soft skills, and a bit of luck. But as their companies grow, they gradually spend less time coding, debugging, or doing deep design.

Instead, they act more like highly technical product managers. They help VPs plan and write high-level product requirements sprinkled with technical terms. They draw boxes on whiteboards and create pretty slides. They write polished documents that keep their leadership happy, and they are either in meetings or on their way to the next one. When they have a technical idea, they dispatch a team to test it out.

Naturally, they still feel deeply technical, until the day they have to resolve a production issue, pass a technical interview, or write extensive code. That is when they realize their skills have grown rusty.

I point this out not to criticize, but to highlight a genuine career challenge. As an engineer, I would rather hone my technical skills. Yet, if you want to climb the corporate ladder, you have to take on more organizational work. The only solution I can think of is to become more like a researcher or a professor. Over the years, good professors spend less time writing papers or deriving formulas. However, their insights are so deep that they still produce amazing results by advising PhD students. But that path is much easier said than done.