Hey OpenAI community,
I’d like to formally suggest that OpenAI consider open-sourcing the weights of text-davinci-003.
Why it makes sense:
-
The model was fully deprecated in January 2024 — zero active commercial use
-
It has been superseded by GPT-3.5, GPT-4, and GPT-4o — no competitive risk whatsoever
-
It is one of the first large-scale RLHF-tuned models ever — massive historical and research value
-
OpenAI already set the precedent with gpt-oss — legacy weights are a natural next step
-
xAI open-sourced Grok-1 with great community reception — this would do the same
Releasing text-davinci-003 would allow researchers to study RLHF alignment at scale on a historically significant model — something that simply isn’t possible today with any publicly available model of this era.
This aligns directly with OpenAI’s founding mission of ensuring AI benefits all of humanity. The model is dead commercially — let the research community give it a second life.
Would love to hear thoughts from the community and any OpenAI staff on whether this is feasible. Anyone else in support of this?