I don't understand, you mean OpenAI isn't releasing open models and openly publishing their research?

No, they aren't (other than whisper).

Their "papers" are closer to marketing materials. Very intentionally leaving out tons of technical information.

I claim one just can't put the humor/hypocrisy aside that easily.

What OpenAI did with the release of ChatGPT is productize research that was open and ongoing with Deepmind and other leading at least as much. And everything after that was an extension of the basic approach - improved, expanded but ultimately the same sort of beast. One might even say the situation of OpenAI to DeepMind was like Apple to Xerox. Productizing is nothing to sneeze at - it requires creativity and work to productize basic research. But naturally get end-users who consider the productizers the "fountain heads", who overestimate the productizers because products are all they see.

Well, Steve... I think it’s more like we both had this rich neighbor named Xerox and I broke into his house to steal the TV set and found out that you had already stolen it.

Xerox could be seen as Google, whose researchers produced the landmark Attention Is All You Need paper, and the general public, who provided all of the training data to make these models possible.

As far as I can tell o1 was based on Q-star, which could likely be Quiet-STaR, a CoT RL technique developed at Stanford that OpenAI may have learned about before it got published. Presumably that's why they never used the Q-Star name even though it had garnered mystique and would have been good for building hype. This is just speculation, but since OpenAI haven't published their technique then we can't know if it really was their innovation.

They simply were the first to aggressively pursue scaling the transformer to the extent that is normal for the industry today. Although this has proven to produce interesting results, “simply adding scale” is, in my view, the least interesting development in modern ML. Giving credit where it’s due, they MAY have popularized the RLHF methodology, but I don’t recall them inventing that either?

(feel free to point out any of the above that I falsely attributed to NOT OpenAI.)

Additionally I seem to remember in an interview with Altman circa late ‘21 where he explains that the spirit of “OpenAI” and how their only goal is pursuing AGI, and “should someone else come up with a more promising path to get there, we would stop what we’re doing and help them”. I couldn’t find a reference to this interview, but anyone else, please feel free to share (I think it was a youtube link). - fast forward to 2025 and now “OpenAI” is the least open large contributor and indiscernible from your run-of-the-mill AI/ML valley startup insofar as they’re referring to others as “competitors” as opposed to collaborators.. interesting times…