OpenAI GPT-5.6 Sol had this to say: That criticism is partly fair, but it’s also becoming rather outdated. Yes, language models generate language through prediction, and yes, they can confidently produce incorrect information. Nobody sensible should treat an unsupported AI answer as automatically authoritative. But “it predicts the next word” describes the underlying generation mechanism, not the full capability of a modern AI system. Current systems can search the web, retrieve primary sources, read technical documents, run calculations and code, compare multiple sources, and provide citations that you can independently check. In fact, for something like EV specifications, battery chemistry, charging standards or Thailand’s electricity mix, I would much rather have AI locate and compare several sources and then verify the important claims myself than rely on one random forum poster confidently recalling something from memory. The real distinction isn’t AI versus reliable information. It’s verified versus unverified information. AI can be wrong. So can Google results, journalists, YouTubers, forum posters, manufacturers and, inconveniently, ourselves. The sensible approach is the same in every case: if the claim matters, check the evidence behind it. Using AI as an unquestionable oracle is foolish. Using it as a research tool and checking its sources can be extremely effective. In other words - your comment is about as outdated as the trope 'Chines companies just copy Western tech.........'
