@atoponce That's funny given that ChatGPT 4o may be linked internally to some math engine.

At least that's my hypothesis. v3 (3.5?) failed miserably (as expected!) when I asked it to compute the inflection points of some curve (explanations correct, computations hallucinated), but ChatGPT+ using the Wolfram Plugin did it just fine.

v4o solved that same exercise perfectly without any (apparent) plugins.

Maybe this "problem" was too easy and it didn't make it to the math engine? (If it has one)

@atoponce

Must be correct!

@atoponce

4o is still experimental. Like all things, AI has it's strengths and it's weaknesses. AI isn't the end all be all, and of course a screwdriver is awful at hammering nails.

@atoponce

A story of dot and comma 😅 .

@atoponce guessing it's thinking like version numbers such as for applications eg python 3.14 but still the primary context should be normal numbers! I think all AI models are doing this from what I've been googling

@atoponce 😁

@atoponce not that I want to be defending AI or anything, but this is kinda like calling a hammer garbage because it cannot drive in screws. You’re obviously not using it for what it was designed for. Arithmetic is not it.

Generative AI does some stuff well. Making shit up being one of its absolute strengths 👍 use it for nonsense and you have a great tool right there