Let me tell you all a little story. Sometime around a or so year ago i reached out to an openai staffer who will not be named who had implied they would be very interested in doing some things open source So, as any good representative of open source, i went to all the people i knew working in open source, and asked them all what things could openai do opensource that they would like to see. I collected maybe 15 bullet points, and gave that list to the staffer. 6 months goes by with no response. So i say whats up lets go? And nothing. So i emailed sama himself. And ya know what’s even? He responded! And setup a meeting with a high level research director. We had an hour long conversation where i went through every item on the list, and he mostly shot every single one down. Mostly for not aligning with their business goals, some with hmm we will consider its. Said if they come up with any ideas they’d reach back out. This was like 6 months ago now, and ive never heard back. Since then, i simply gave up trying with them. I gave them all the reasons i could as to why they would benefit from it, but went on what at least at the time was deaf ears. Here is that list again @sama since you seem to have finally seen what i was thinking, thanks at least for answering the email. - Open source your models or at least legacy and deprecated models - Research papers on what has worked for OAI, GPT-4 information such as architecture etc - Opensource datasets for IFT/RLHF/Toolformers/Plugins/FunctionCalling/Web Browsing? (Addendum, reasoning) - Provide preferential access channels for vetted OSAI groups. - Vet internal research for release, with prerogative to release as much as possible, even if on a delay commensurate with business goals and safety concerns. - Have official townhall type meetings with major open source projects/teams - Tools to filter data - Contribute directly to public open source AI projects. - Plug & Play RLHF training code/Reward Models - Alternatives to Text Models (Voice/Music/Image/Video) (preferably OS) - Open sourced moderation classifier model/Classification Model's dataset - Base gpt-4 model access for researchers - opt in to removing the gpt4 moderation - Release roadmap - Opening up the ToS to make it unambiguous that training models on openai outputs are all good.