I suspect that's why OpenAI recently added the share conversation thing, people are just looking at 3.5 and scoffing at it as it fails at things 4 would do just fine, and then assume it applies to all models. They've got a marketing problem that can't really be solved without making 4 public or showing people volumes of examples of what it can do. I was pretty convinced by the launch demo, seeing it not make the same mistakes I've seen 3.5 do when using it, but basically nobody's seen that one.