3 ms·
In the thread yesterday it was brought up that if you use the API it feels considerably less hamstrung than the ChatGPT client, and this tweet seems to fit the
by mustacheemperor 3y ago
In the thread yesterday it was brought up that if you use the API it feels considerably less hamstrung than the ChatGPT client, and this tweet seems to fit the assumption that the ChatGPT product is being tuned or governed differently from the API.
>The API does not just change without us telling you. The models are static there.
This reads to me as specifically indicating the models are not static elsewhere, ie, in ChatGPT.
- jonplackett 3y agoIt feels less hamstrung on output, but is instead hamstrung by being incredibly slow. GPT-4 via API will sometimes take 30 seconds + to respond to simple questions without any chat history, where through Chat GPT it will give you near enough instant replies only slightly slower than 3.5 turbo.
- danielbln 3y agoMaybe the reverse is true? To provide faster performance they dumb down gpt4 (quantization?) that serves the chatgpt UI but the API is full power albeit slow.
- jonplackett 3y agoI think it’s just prioritising ChatGPT. The reason that’s my guess is that the API is occasionally much faster and then goes slow again. It’s a bit all over the place which leads me to suspect it’s based on demand