Nowadays you can explain a 10 step algorithm to chatgpt and it will do the math for you.
Good luck trying to calculate hash functions step-by-step with Chat GPT (it won't be even close to results, which you can get on
sha256algorithm, and it will just hallucinate some constants out of
Wikipedia description, but nothing beyond that). Or try asking about
Schoof-Elkies-Atkin algorithm on real elliptic curves (it will repeat, that this topic is too complex, because many PowerPoints said that; or it will hallucinate, and tell you completely wrong results, even if all numbers will be below 100). Existing AI models are really bad, when it comes to math. They are nowhere close to tools like
Sage or
WolframAlpha.
You are the expert.
If you say we are good, then I can’t argue. J-Lopp clearly is not in the same boat.
A decade ago all people had was google and when they decide to build a very complex algorithm or do a complicated math problem, they were on their own.
Nowadays you can explain a 10 step algorithm to chatgpt and it will do the math for you.
Well from what I've seen from AI, I wouldn't call it intelligence. It is just database of whatever information that existed and its devs included in its database. In other words it is not capable of innovation. It just provides you with solutions that already exist. For example if you ask it to help you solve ECDLP, it will give you the information that already exists in its database which is something you could have found on google 10 years ago as well as today.
AI is also not improving computation speed which is what's needed to solve ECDLP.
How do we know that what they have shown us so far is all they got?