← Back to context

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....

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.

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.