Re:How 'Open' Is OpenAI, Really?
by fenomas on Monday March 13, 2023 @01:37AM
(#35131543)
Question for someone who follows this space: are there any significant models/tech/etc that OpenAI has made open? Either before or after they changed business models. What they've kept private is well-known, but are there any checkboxes in the other column?
[ Parent
| on HN ]
Re:How 'Open' Is OpenAI, Really?
by scottydog51834 on Monday March 13, 2023 @01:41AM
(#35131569)
GPT-2
Here is one cool (and recent) research project using GPT-2 model weights.
https://www.lesswrong.com/posts/cgqh99SHsCv3jJYDS/we-found-a...
TFA claims GPT-2 isn't open. Is it that the trained weights are open but the project that did the training isn't?
Re:How 'Open' Is OpenAI, Really?
by alwayslikethis on Monday March 13, 2023 @03:53AM
(#35132466)
Realistically the most difficult part of training a LLM is curating a good data set. OpenAI never published their training data.
Re:How 'Open' Is OpenAI, Really?
by polygamous_bat on Monday March 13, 2023 @01:40AM
(#35131558)
Off the top of my head, CLIP the vision-language model, and Whisper, the audio to text model, have their trained weights and the architectures open. They did not open source the training dataset, although I fear it may be too much of an ask given all the possible legal issues.
whisper
Whisper. Speech recognition and translation. Probably the best general-purpose speech recognition available right now. Certainly the best you can run on your own hardware. Open-sourced code and publicly free and available models. https://github.com/openai/whisper