

I’ve seen a Claude screenshot posted in a community of computer scientists where it “proved” P=NP through circular reference and cursed mathematics.
For context, P and NP are sets of problems. If an algorithm exists where the time it takes to determine if a solution to a problem is correct scales polinomially relative to its input size, it’s in NP. If the time to solve the problem scales polinomially, it’s also in P, a subset of NP.
The question now is if P = NP: Is every problem we can verify in polinomial time also solvable in polinomial time? We just don’t know, although we think it probably isn’t. But that’s just a guess.
So someone asked Claude and it came up with: https://fixupx.com/connacher_/status/2079541927553212471 which, to a maths person, is hillariously incorrect.
I just don’t trust in AI’s ability to reliably produce true statements.



Congrats on making it, something like this can be quite hard.
By the way: You should probably move from Telegram to something like Signal to get strong end-to-end encryption by default back. Telegram’s secret chats aren’t the default, are inconvenient to set up and cryptographically strange.