GPT-4 is the worst model in terms of misinformation, yet it's still the default model

Been lurking here for a while, been reading and agreeing with the complaints about the misinformation in GPT-4 and GPT-4 turbo. That led me to try out the older models (GPT-3.5, GPT-3) and I was pleasantly surprised. It is night and day. GPT-3.5 turbo for instance is much more accurate, often more informative and more willing to admit when it doesn't know something. I can't even be bothered to compare them to GPT-4 turbo, the margins are so big. Especially with factual information. It's even good with historical information, which is something I'm very good at and can always catch bs with. So you can imagine how even more glaringly bad the misinformation is in GPT-4 for me.<br><br>I can't even be bothered to compare them to GPT-4, the margins are so big. Especially with factual information. It's even good with historical information, which is something I'm very good at and can always catch bs with. Especially with factual information. It's even good with historical information, which is something I'm very good at and can always catch bs with.<br><br>It's even good with historical information, which is something I'm very good at and can always catch bs with.