Live data from Hacker News

Terence Tao: At the Erdos problem website, AI assistance now becoming routine

mathstodon.xyz

31–40 of 96 posts

Re: Terence Tao: At the Erdos problem website, AI assistance now becoming routine

#31
post #16

I hope we continue to see gains for scientific professionals and companies doing research. Even imperfect assistants increase leverage.

iOS formalization app mentioned by Tao (beta)

https://aristotle.harmonic.fun/

TIL: startup founded by Robin Hood CEO

Re: Terence Tao: At the Erdos problem website, AI assistance now becoming routine

#32
post #27

Having the ability to throw math heavy ML papers at the assistants and get simplified explanations / pseudocode back is absolutely amazing, as someone who's forgot most of what I learned in uni, 25+ years back and never really used it since.

This is where LLMs shine for learning imo: throwing a paper in Claude and getting an overview then being able to ask questions. Especially for fields that I didn’t study at the Bachelors or Masters level, like biology. Getting to engage with deeper research with a knowledgeable tutor assistant has enabled me to go deeper than I otherwise could.

I'm not sure the exact dollar value of feeling safe enough to ask really stupid questions that I should already know the answer to and I'd be embarrassed if anyone saw me ask Claude, but it's more than I'm paying them. Maybe that's the enshittification play. Extra $20/month if you don't want it to sound judgey about your shit.

Re: Terence Tao: At the Erdos problem website, AI assistance now becoming routine

#33
post #24

Erdős I was told by a hungarian, that hungarian written spelling and spoken pronunciation is pretty precisely aligned compared to, say, english. Except when it comes to names when it gets a bit random! Why not do the bloke the decency to spell his name correctly? Those diacritics are important. Anyway, I was told that Paul's name is very roughly pronounced by an anglophone as: "airdish".

I could be wrong, but FWIW I doubt Hungarians include diacritics that don’t exist in Hungarian (like ñ) when writing foreign names.

Re: Terence Tao: At the Erdos problem website, AI assistance now becoming routine

#34
post #27

Having the ability to throw math heavy ML papers at the assistants and get simplified explanations / pseudocode back is absolutely amazing, as someone who's forgot most of what I learned in uni, 25+ years back and never really used it since.

This is where LLMs shine for learning imo: throwing a paper in Claude and getting an overview then being able to ask questions. Especially for fields that I didn’t study at the Bachelors or Masters level, like biology. Getting to engage with deeper research with a knowledgeable tutor assistant has enabled me to go deeper than I otherwise could.

Isn't there a risk that you're engaging with an inaccurate summarization? At some point inaccurate information is worse than no information.

Perhaps in low stakes situations it could at least guarantee some entertainment value. Though I worry that folks will get into high stakes situations without the tools to distinguish facts from smoothly worded slop.

Re: Terence Tao: At the Erdos problem website, AI assistance now becoming routine

#35
post #27

Earlier quoted context omitted.

This is where LLMs shine for learning imo: throwing a paper in Claude and getting an overview then being able to ask questions. Especially for fields that I didn’t study at the Bachelors or Masters level, like biology. Getting to engage with deeper research with a knowledgeable tutor assistant has enabled me to go deeper than I otherwise could.

Isn't there a risk that you're engaging with an inaccurate summarization? At some point inaccurate information is worse than no information. Perhaps in low stakes situations it could at least guarantee some entertainment value. Though I worry that folks will get into high stakes situations without the tools to distinguish facts from smoothly worded slop.

I've been doing this a fair amount recently, and way I manage it is: first, give the LLM the PDF and ask it to summarize + provide high-level reading points. Then read the paper with that context to verify details, and while doing so, ask the LLM follow-up questions (very helpful for topics I'm less familiar with). Typically, everything is either directly in the original paper or verifiable on the internet, so if something feels off then I'll dig into it. Through the course of ~20 papers, I've run into one or two erroneous statements made by the LLM.

To your point, it would be easy to accidentally accept things as true (especially the more subjective "why" things), but the hit rate is good enough that I'm still getting tons of value through this approach. With respect to mistakes, it's honestly not that different from learning something wrong from a friend or a teacher, which, frankly, happens all the time. So it pretty much comes down to the individual person's skepticism and desire for deep understanding, which usually will reveal such falsehoods.

Re: Terence Tao: At the Erdos problem website, AI assistance now becoming routine

#36
I was driving around tonight doing errands and while I was doing so, had a great conversation with chatgpt as to the intimate details of the llvm and GCC pipeline schedulers. It's a huge productivity boost. It has now taken notes for me for some compiler stuff I'm experimenting with. This would previously have been impossible.

Re: Terence Tao: At the Erdos problem website, AI assistance now becoming routine

#37
post #27

Earlier quoted context omitted.

This is where LLMs shine for learning imo: throwing a paper in Claude and getting an overview then being able to ask questions. Especially for fields that I didn’t study at the Bachelors or Masters level, like biology. Getting to engage with deeper research with a knowledgeable tutor assistant has enabled me to go deeper than I otherwise could.

Isn't there a risk that you're engaging with an inaccurate summarization? At some point inaccurate information is worse than no information. Perhaps in low stakes situations it could at least guarantee some entertainment value. Though I worry that folks will get into high stakes situations without the tools to distinguish facts from smoothly worded slop.

There is, but just ask it to cite the foundational material. A huge issue with reading papers in topics you don't know about is that you lack the prerequisite knowledge and without a professor in that field, it may be difficult to really build that. Chat GPT is a huge productivity boost. Just ask it to cite references and read those.

Re: Terence Tao: At the Erdos problem website, AI assistance now becoming routine

#38
post #24

Erdős I was told by a hungarian, that hungarian written spelling and spoken pronunciation is pretty precisely aligned compared to, say, english. Except when it comes to names when it gets a bit random! Why not do the bloke the decency to spell his name correctly? Those diacritics are important. Anyway, I was told that Paul's name is very roughly pronounced by an anglophone as: "airdish".

I could be wrong, but FWIW I doubt Hungarians include diacritics that don’t exist in Hungarian (like ñ) when writing foreign names.

You are wrong.

Re: Terence Tao: At the Erdos problem website, AI assistance now becoming routine

#39

'Vibe formalizing' is a logical extension of 'vibe engineering' implemented by 'vibe coding'. Sometimes I have trouble with getting the individual puzzle pieces of a problem to fall into place, where a hypothetical 'Move 37 As A Service' to unify informal methods with mathematical rigor deserves to be explored!

I had put on the back burner some polyhedral compilation papers. I had read all the material, but some key questions still meant it was not possible for me to implement it. In particular, I was looking at barvinoks counting algorithm and did not understand why you needed to expand the polynomials in a pointed cone. However, chatgpt correctly led me through the reasoning. Could it have made a mistake? Of course. And it did. However, since my confusion meant that I was also wrong, bouncing the idea back and forth was really useful. Plus the ai bots are better at understanding your own particular points of confusion.

Re: Terence Tao: At the Erdos problem website, AI assistance now becoming routine

#40
post #22
post #10

Earlier quoted context omitted.

Because he is smart enough to use the existing (frontier) tools to get good results and create a sort of collaborative environment that is novel for research maths.

Indeed. Who’s to say whether Wiles would have used AI assistance if it had, you know, existed, in 1994.

Wiles's initial presentation of the proof had a serious flaw that killed the whole thing until he found a workaround. I don't remember how long it took him to get out of the jam, but I'm sure he would have handed his credit card to the Devil himself if he thought it might help.

People who don't take advantage of the best available tools and techniques don't get to that level to begin with.

Post reply on HN