‘Basically zero, garbage’: Renowned mathematician Joel David Hamkins declares AI Models useless for solving math. Here’s why
https://m.economictimes.com/news/new-updates/basically-zero-garbage-renowned-mathematician-joel-david-hamkins-declares-ai-models-useless-for-solving-math-heres-why/articleshow/126365871.cms
12 Comments
Holy cow. A next word predicting machine can’t do simple math.
Frankly I am shocked.
Edit: This is a Joke. JOKE.
The article is self repeating every two sentences reading this felt like having a stroke
There’s a reason LLMs generate code in order to do math based on data.
Real old-man-yelling-at-clouds energy here. AI has already been used at high levels to accelerate math research, and it’s only going to get better.
Wait I thought it can solve the world math olympiad better than almost any human alive.
Strong claim, but it feels a bit overstated. Current LLMs are bad at deep, novel mathematical reasoning especially proofs, but they’re quite useful for intuition building, checking steps, exploring examples and catching errors. They’re tools, not mathematicians and like calculators at first their value depends on how they’re used.
For now maybe, but AI will soon likely be outperforming humans in reasoning and thinking skills. But unfortunately, this will happen not by AI becoming significantly smarter or more powerful, but just relatively as humans becoming more and more stupid due to a whole generation of society developing cognitive atrophy as a result of outsourcing all their high level thinking and reasoning skills to AI.
Can he explain how the latest models got gold on the International Math Olympiad?
I tested Claude and Deepseek with some composite integrals to solve and the results were actually correct and very well explained. So as a noob like myself who vaguely remembers some first year of university maths, they are not useless at all !
That article was written by AI.
This article is dogshit, it’s a single paragraph about some random fuckass without even evidence he said what it claims or why he thinks that way.
LLMs are not machine learning AI in general.
I don’t see how models that approach the statistical most likely answer wouldn’t be able to apply this to math. It did work in 2015 with Go already, which is basically “math”.