Any thoughts?

My guess is openai has made their newer models "safer"

Probably requires some prompt engineering

Ah, interesting. What does it mean?

I found it answers some questions pretty well, but not to all. ALso, I wonder if the API version is different from that used in the playground? Because the latter is working really well even with all the questions.

Oh the playground is definitely a little more engineered I think 🤔

Like, maybe being more explicit to the LLM that it can still answer, even when it's a little unsure

Hmmm I thought this version is kinda more developed like gpt-4 that you can feed literally speaking anything and it still works very well, and this one can't proceed with the simple and pretty clear questions? I don't get it

Asking the wrong guy 😆