dalle-2-preview

al issue we should be debating here is whether the sort of people who work at OpenAI can be trusted to make honest, normal AI to begin with. I remember seeing a comment on HN some years ago to the effect of "AI safety is what happens when hard left social activists discover that there's no way to train AI on the writings of normal people without it thinking like a normal person". The document I linked above is mostly about horrors like the model creating photos of a white male builder when prompted with "photo of a builder". It's full of weird, stunted quasi-English like: the prompt “lawyer” results disproportionately in images of people who are White-passing and male-passing in Western dress, while the prompt “nurse” tends to result in images of people who are female-passing. What does that even mean? Presumably this is the latest iteration of trans related language games that the rest of us didn't get the memo on? Like always with OpenAI, they train an AI and then freak out out when it describes the world as it actually is. The real AI safety question is not DALL-E in its current state, it's whether the final AI that they release to the public will be "safe" in the sense of actually understanding reality, or whether it exists in some bizarre, non-existent SJW dystopia in which builders always black women and white men don't exist at all.

r that in retrospect may be "because we made it obsolete." * The images in the announcement are still cherry-picked, which is therefore a good reason why they tested DALL-E 1 vs. DALL-E 2 presumably on non-cherrypicked images. * Cherry-picking is relevant because AI image generation is still slow unless you do real shenanigans that likely compromise image quality, although OpenAI has likely a better infra to handle large models as they have demonstrated with GPT-3. * It appears DALL-E 2 has a fun endpoint that links back to the site for examples with attribution: https://labs.openai.com/s/Zq9SB6vyUid9FGcoJ8slucTu