Each release from openAI gives me less hope for them and this whole AI boom. They should be leading the charge of highlighting how the current generation of LLMs fail, not churning out half-baked overhyped products.
Yes, they can do some cool tricks, and tool calling is fun. No one should trust the output of these models, though. The hallucinations are bad, and my experience with the "reasoning" models is that as soon as they fuck up (they always do) they go off the rails worse than the base LLMs.
Yes, they can do some cool tricks, and tool calling is fun. No one should trust the output of these models, though. The hallucinations are bad, and my experience with the "reasoning" models is that as soon as they fuck up (they always do) they go off the rails worse than the base LLMs.