Post #620

> We're seeing the first signs of 1000x acceleration in science and tech progress, driven by Al Al is now solving extremely hard math problems (almost daily) that humans couldn't solve until this moment (…) GPT 5.2 Pro develops faster 5x5 circular matrix multiplication algorithm (…) «I have little doubt that AI tools will substantially change the way mathematics is done, likely pretty soon But communicators like this are doing their audiences a disservice In this particular example the “new” contribution from the AI tool was plugging (n,m)=(5,3) into the expression 2n-m. I’m not exaggerating The “number of multiplications” in the result in question was shown by Winograd ~45 years ago to be 2n-m, where the matrices in question are nxn, and m is the number of factors of the polynomial x^n-1 The “result” here is that this polynomial has 3 factors over Q(sqrt(5)), and then that 2*5-3=7 That’s it Winograd’s result itself is an application of the Chinese Remainder Theorem (…) That automated tools can make progress on minor open questions is incredibly impressive But that doesn’t excuse inaccuracy about current capabilities» https://x.com/littmath/status/2013037012765331883