On February 23, Google stopped allowing their new Gemini model to generate images containing people. When will they re-enable this ability?
Resolves to the time period when Google officially re-enables this ability. If people discover a jailbreak that causes the current model to produce images of people, that doesn't count.
1,000
3.00@0482 I guess because this is a linked/dependent multiple choice market, it's not possible to resolve individual options, but I'm not sure.
@ML12d1 Agreed, I've tried to resolve the already-passed dates to NO, but don't see any way to resolve some of the answers without closing the market. If someone sees one, please let me know!
Not trying to troll but I can't actually think of a clear set of rules for"representation* which would be self consistent. One reason is that these systems are usually judged on single images, so being statistically accurate won't work. For example, for teams of four programmers, if the industry is 1/4 female, women won't be equally distributed, 1 per team. There may be teams with 0 or 2 or 3. Just by chance sdone teams will be all men. Yet, no one would fight back if an image of four male programmers were being criticized.
Secondly, historical representation has similar problems.
AI, given that Google will probably attempt to be "beyond criticism" in their confidence in the doctrine of representation, and that it's basically impossible to do it well without overdoing it, I don't see a way out. There really is no solution which satisfies all parties.
OpenAI has a less strong but still invasive solution where they irresistibly insert us-style diversity into any semi modern or later image with more than three people, but leave others alone. They can defend themselves by highlighting times it turns on, but can be attacked for when it doesn't.