Model documentation or request for a "model" category?

Hello,

Under “Playground”, we can find a list of available models including “gpt-4-turbo-2024-04-09”. However, where’s the documentation that explains each model? For instance, for this model, it’s likely that it was made available on that date, but then when was its training data’s cut-off date?
Then, for “gpt-4-turbo-preview”, since it’s listed at the very top, probably it’s the newest, so, how’s it different from “gpt-4-turbo” and in what way?

Could we have a category named “model” to specifically provide information about each model?

Hi - you can get started here with an overview: https://platform.openai.com/docs/models

On the OpenAI website, you can also find additional information. For example see details on GPT-4 here: https://openai.com/index/gpt-4/

1 Like

A good many model names are pointers, stable names that are aliases to the real versioned models:

  {"id": "gpt-3.5-turbo", "alias_to": "gpt-3.5-turbo-0125"},
  {"id": "gpt-3.5-turbo-16k", "alias_to": "gpt-3.5-turbo-16k-0613"},
  {"id": "gpt-4", "alias_to": "gpt-4-0613"},
  {"id": "gpt-4-32k", "alias_to": "gpt-4-32k-0613"},
  {"id": "gpt-4-turbo", "alias_to": "gpt-4-turbo-2024-04-09"},
  {"id": "gpt-4-turbo-preview", "alias_to": "gpt-4-0125-preview"},
  {"id": "gpt-4-vision-preview", "alias_to": "gpt-4-1106-vision-preview"},

0301, 0314: Snapshot models from March 2023 with little alteration since then - highly skilled but can also hallucinate facts.

0613: full models that can use functions. Models that continue to be updated (having the original release back would be a godsend to devs).

others: post Devday models where GPT-4-Turbo is significantly reduced in cost and quality, favoring safe trained chatting to enhance the efficiencies made in computation, and denials. The final gpt-4-turbo-2024-04-09 which has vision can write pretty good code again - if you like stuff you didn’t want being inserted and rewritten and an AI that gets stuck, unable to improve and forgetting and simplifying after a conversation grows.

2 Likes

and @jr.2509
for “moderations” models, I wonder what algorithm that OpenAI uses for classification and its training data size. In my experience, deep learning does not necessarily produce best results even with a decently large training data set.