> I think most of the other gains we've seen since are related to fine-tuning on intentional application-specific tasks. IMO the only real "significant" improve

> I think most of the other gains we've seen since are related to fine-tuning on intentional application-specific tasks. IMO the only real "significant" improvement is the native multi-modal models. Let me say, as someone with no connection to OpenAI, that Deep Research mode is an absolute game-changer for my use and well worth the money. Obviously, YMMV, but for really does do deep research and organize it all in an excellent way. I haven't yet noticed an error in its output when researching general subjects. It doesn't solve the problem of not necessarily using code examples from the relevant version of a Rust library, so it has a good ways to go for Rust coding. But I am very, very impressed with its usefulness for general research. My point is only, if you haven't tried it, don't assume that there isn't already a game-changer out there for many uses.