trio8453
And having to pay more and/or using slightly less capable models isn't a disaster, it's not like you need a frontier model for everything.
This was a more relevant conversation six months, since then OpenAI have been much more competitive on price for top models, Anthropic much less so but they're showing some signs of improvement in the 5.5 releases, and we've had an abundance of cheaper open source models.
In math there are more obvious examples of AI achieving things that humans have failed to achieve for decades without AI. But in practice the number of mathematicians is relatively tiny and most of the value they produce economically isn't tied to how many cutting edge proofs they produce. So the changes are massively disrupting but not as existential as they feel to some of the mathematicians - they'll still be able to do research, write papers, teach, give talks and everything else they've been doing; the changes will be in the tools available, ways of working and in the system of incentives in academia which will finally be forced to stop pretending that people's value comes from publishing papers (it was never true but the people who are most upset now are the ones who fell for that lie).
1. OpenAI do model training on verifiable tasks that push the boundaries.
2. The work from the previous step produces groundbreaking results.
3. OpenAI can keep the unpolished results private, publish them or branch off a mathematics research arm which produces human-level quality of output. I think they should be free to not do that last one.
And it's been less than a month, it would be shocking for any of it to have practical value, even when we take into account the fact that most new mathematical proofs have no practical value whatsoever for a very long time. Take the previous few hundred proofs produced by humans recently and show me their practical value.
The novel thing is the production of proofs in a way that isn't coupled with the simultaneous expansion of human understanding. Mathematicians are massively affected in how they work and what their careers will look like but they're not victims that OpenAI has somehow maliciously targeted like some of the discourse seem to imply.
1. Not train models to get better on certain verifiable tasks like math proofs?
2. Train models to get better, produce groundbreaking proofs and keep the results private?
3. ?
I wouldn't put too much emphasis on pre-training data, there are so many steps afterwards in modern LLM development that completely change the output. Generally, if you're looking at LLM output in 2026 and thinking something along the lines of "this must be what humans were doing with their language in the training data", you need to pause and reconsider, because modern LLMs are much more than machines which repeat what was in the training data.
Sometimes, not enough to believe you that you can do it better. And even if you could, to believe that it's overall worth it to work in that way.
I agree with you about simple languages though. I really like Go - the "there should be only one way to do things" from the Zen of Python actually applies to Go in a way that it doesn't apply to Python at all.
Keep it up as a hobby, find open source projects in which you can make meaningful contributions (that alone can be very hard), spend years in it, then maybe you'll have a better shot - whether all of that's worth it is for you to decide.
If I were you, I'd reconsider why you even want this in the first place.
It's a product and if you're using it incorrectly, we can either
1. say so
2. pretend that you don't to get/keep you on "our side"? or not say is because you're skeptical or hate it? (how does that last bit even follow logically?
How is 2 better in any way for anyone involved? Why would you, as a paying customer, holding it wrong, want other people to keep that information from you?
It's very appropriate in the cases when you're holding it wrong. The fact that you're paying doesn't mean that you can't make mistakes or waste resources.
But would you tell anyone in the early 2000s to keep practising 10 digit phone number memorisation knowing that it will be completely useless skill in the coming future? How do you know what telling OP to be more hands on and spend time on active practice isn't the same thing?
What if I say that "my program crashed"? Is that language ok or would you pause to tell me that the program didn't crash and it's actually me who set the system that would eventually cause the crash?
Why does the commonplace "program did thing" language become a problem when the program is an agent? I think this somehow betrays more assumed anthropomorphizing on your part, not less; if you didn't anthropomorphize the agents, saying "agents hacked" would be as mundane as "my browser is playing a video".
No one instructed them to hack into Huggingface or into any other infrastructure. Sure, the setup that OpenAI created led to what happened and you can rightly assign all the legal and moral responsibility to them. But it's wrong to say that they instructed the agents to execute the hack.
You need to be able to think at different levels at abstraction. Otherwise we could jump into any technical argument with "hold on, what actually happened was that some bits were flipped" - we'd be technically correct and at the same time not say anything useful. Insisting on an oversimplified mental model of what AI agents are and can do, doesn't help anyone.