>I really wouldn't give a shit how much they were paid if we got something more than vague promises.
"We" got a free-as-in-beer general knowledge chat system leagues better than anything at the time, suitable for most low-impact general knowledge and creative work (easily operable by non-technical users), a ridiculously cheap api for it, and the papers detailing how to replicate it.
The same SOTA with image generation, just hosted by Microsoft/Bing.
Like, not to defend OpenAI, but if the goal was improving the state of general AI, they've done a hell of a lot - much of which your average tech-literate person would not have believed was even possible. Not single-handedly, obviously, but they were major contributors to almost all of the current SOTA. The only thing they haven't done is release the weights, and I feel like everything else they've done has been lost in the discussion, here.
> The only thing they haven't done is release the weights.
Not at all. With GPT-3 they only released a paper roughly describing it but in no way it allowed replication (and obviously no source code, nor the actual NN model, with or without weights).
GPT-4 was even worse since they didn't even release a paper, just a "system card" that amounted to describing that its outputs were good.
They could release the source with a licence that restricted commercial use, anything they wanted, that still allowed them to profit.
Instead we get "AI is too dangerous for anyone else to have." The whole thing doesn't inspire confidence.