Post

I've seen a few negative reactions to my recent vintage mods using AI, and I get it. The tech world is changing rapidly and it's frightening, even for someone who has expected this. I've had existential dread since I first used Claude Code last year Someone on X put it pretty well:

Yes, the status quo in software is dead. As scary as it seems now, it's possible that in a few years these tools will seem commonplace and maybe even boring You'll be able to run coding models equivalent to Opus 5.5 locally on a GPU in your device, so you won't need to "boil oceans" to do it

We are in an early period where demand for compute is outstripping supply, and it's leading to a speculative build-out of datacenters fueled by Big Tech billions that are delivering more capable AI a little sooner, but at a precarious cost I believe locally-run AI will level that out eventually

Also, some people have the mistaken impression that AI agents aren't very capable, based on the shortcomings of AI models as recently as January. But capabilities have improved rapidly, and they are not slowing down In the mean time, I am going to keep experimenting and showing you results

Their capabilities only ever seem good when they're conducting tasks outside of the user's specific wheelhouse, no matter what their skillset may be, which suggests to me that no, they're not all that capable. And they are not worth the massive human health, energy and env. costs to play with.

I feel like if this were true then the entire field of mathematics would not currently be having a prolonged, public, existential crisis about what to do now that the computers can prove theorems just as well as them

It only proves theorems when it steals someone else's work www.forbes.com/sites/sandyc...

That would be the same dispute in which the original blog post from Alpöge and Buckmaster explicitly enumerated the LLMs they used in the process, yes cims.nyu.edu/~tristanb/st... The question in dispute there is “did OAI illicitly access A&B’s work for their own gain”, not “can LLMs do maths”

the worlds best mathematicians don't believe "it can't produce original mathematical work, only steal existing work" so you should believe them instead of random people on the internet

people whose pipeline into the mathematical world is forbes magazine will form some pretty bizarre views but you can just read mathematicians on twitter and here talking about this, now