@bcongdon This looks like it can clearly resolve NO based on the internal and external branding of text-davinci-XXX as a refinement of GPT-3 and ChatGPT as still being "GPT 3.5."

https://twitter.com/sama/status/1607220186146541568

@FranklinBaldo I miss read this tweet

related

Does chatgpt count?.

It seems better at some tasks, though it's hard to tell how much, and it might actually be using diferent models depending on context.

I think there is still going to be a "gpt4" model but that might take longer.

OpenAI says ChatGPT was fine-tuned from a model in the "GPT3.5" series. I'd say it doesn't count.

@wadimiusz The announcement email I got from OpenAI had the subject "New GPT-3 model: text-davinci-003", so I'd say it doesn't count.

@VictorLevoso btw, chatgpt is so bad lol. I'm not impressed. it's significantly worse than t-d-3

@bcongdon That's a different model that released arround the same time.

I think it doesn't count anyway for the reason vadim mentions.

@bcongdon text-davinci-002 isn't GPT-4 at all

@bcongdon text-davinci-003 was announced last night. they are basically viewed as just incremental GPT-3 variants though.