also, for posterity / future historians, here is some (public, non-exhaustive, from my recollection) info on somewhat more obscure historical openai model naming around the 2020-2022 era:
a very useful (non exhaustive) resource is the “Model index for researchers” post which no longer exists but which is thankfully archived: https://archive.is/XDCN0
a cushman model was very briefly made available but then disappeared forever. my memory is a bit fuzzy but i believe this is separate from the code-cushman-001 mentioned in the model index
the first GPT-3 instruction following model was launched as davinci-instruct-beta, subsequent IF models were text-davinci-001 (GPT-3), text-davinci-002 (GPT-3.5), text-davinci-003 (GPT-3.5)
the name GPT-3.5 was not public until the existence of this index. so nobody really knew that text-davinci-002 was no longer GPT-3 until then
also, for posterity / future historians, here is some (public, non-exhaustive, from my recollection) info on somewhat more obscure historical openai model naming around the 2020-2022 era:
a very useful (non exhaustive) resource is the “Model index for researchers” post which no longer exists but which is thankfully archived: https://archive.is/XDCN0
the GPT-3 series was launched as ada, babbage, curie, davinci. davinci was the full size 175B and the others were smaller models (see https://blog.eleuther.ai/gpt3-model-sizes/ )
a cushman model was very briefly made available but then disappeared forever. my memory is a bit fuzzy but i believe this is separate from the code-cushman-001 mentioned in the model index
the first GPT-3 instruction following model was launched as davinci-instruct-beta, subsequent IF models were text-davinci-001 (GPT-3), text-davinci-002 (GPT-3.5), text-davinci-003 (GPT-3.5)
the name GPT-3.5 was not public until the existence of this index. so nobody really knew that text-davinci-002 was no longer GPT-3 until then