News for hackers. Stuff that matters.
Re:The genie escapes: Stanford copies the ChatGPT AI for les... by LASR on Monday March 20, 2023 @05:15PM (#35238871)
In my personal testing, I throw some sophisticated use cases at LLMs - particularly chain of thought reasoning. None of the models out there are able to do this this well, except for the OG GPT-3 Davinci-003. Even the newer turbo models are not as good.

I am playing around with GPT-4 this week though. Let’s see how that goes.

Re:The genie escapes: Stanford copies the ChatGPT AI for les... by stavros on Monday March 20, 2023 @05:20PM (#35238940)
The newer turbo models are the ChatGPT models, and are worse than text-davinci-003, in my experience. The gpt-4 model is also not as good as the GPT-4 chat version, which is very odd.
Re:The genie escapes: Stanford copies the ChatGPT AI for les... by LASR on Monday March 20, 2023 @10:20PM (#35241774)
Yeah I think it lines up with Davinci being 10x as expensive as Turbo.