Comment by whimsicalism
3 months ago
i'm surprised by the lack of sycophancy in o3 https://www.reddit.com/media?url=https%3A%2F%2Fpreview.redd....
3 months ago
i'm surprised by the lack of sycophancy in o3 https://www.reddit.com/media?url=https%3A%2F%2Fpreview.redd....
Well the system prompt is still the same for both models, right?
Kinda points to people at OpenAI using o1/o3/o4 almost exclusively.
That's why nobody noticed how cringe 4o has become
They have different uses. The reasoning models aren't good at multi-turn conversations.
"GPT-4.5" is the best at conversations IMO, but it's slow. It's a lot lazier than o4 though; it likes giving brief overview answers when you want specifics.
people at OAI definitely use AVM which is 4o-based, at least
pretty easy to understand - you pay for o3, whereas GPT-4o is free with a usage cap so they want to keep you engaged and lure you in.