Man… the new 0613 is horrible. It no longer follows directions or steps. If OpenAI wants to be the top-dog, then they have to keep their models stable. Otherwise, it’s impossible for tools integrating these models since they lack stability. I’m having to re-do most of my prompts because GPT 3.5 Turbo is so poor now. I don’t even know if I can because it feels like I’m no longer talking to a teenager, but a baby.
We’re in the same boat brother, everyone who’s even remotely a power user knows what they did. They decreased the intelligence to save money (less computational resources on comprehending prompts), and now that it’s officially released, decreased the speed to that of the prior engine. Basically we all got screwed.
Agreed gpt-3.5-turbo-0613 is horrible vs gpt-3.5-turbo-0301
gpt-3.5-turbo-0613 provides much less relevant and less precise answers. He is much more verbose with no added value with regards to the quality the answer.
I start to regret I have devoted time an efforts building API to query GPT.
I was very much please with gpt-3.5-turbo-0301. I use it with embeddings, the whole in French langage. gpt-3.5-turbo-0613 is almost useless to me now in the same context.
I do not understand why there is so much difference on a sub gpt-3.5-turbo-0"XXX" version.
The work I have done just worth nothing when gpt-3.5-turbo-0301 is to be turned off.
Please keep gpt-3.5-turbo-0301 up. gpt-3.5-turbo-0613 is bad.
Exactly the same here. The update was exactly what I needed (follow system better and more context)
In my case its definitely better at following system, I could remove most workarounds I used to make it keep following system
Not translations though, english