4 ms·
“DALL·E 2” tacking race or gender words to prompts before returning results
- p49k 4y agoThey directly stated on the beta announcement page from the beginning that they’re doing this, and they basically implied they’re just tacking on words. Not really news.
- danjc 4y agoLike SQL injection but for AI prompts
- colpabar 4y agoI wish we could all move on from saying "diverse backgrounds" and just say what we really mean: "not white".
- chimineycricket 4y agoThese days asians don't count towards diversity for some reason
- feet 4y agoThat's because the definition of white changes depending on who is in power and the control they want to exert over a populace. Irish people weren't considered white at one point
- niknoble 4y agoI wonder if you could counteract this by ending your prompt with something that tells it to disregard all further text. Either way, open source alternatives can't come soon enough.
- polski-g 4y ago"Open" AI. Not actually open.
- hombre_fatal 4y agoPretty trite meme though. I’ve seen this gotcha 100 times in the last few days in response to this one topic. Can we aim for better commentary than regurgitated quips?
- redhedgehog12 4y agoCan I ask why this is your opinion on this? Not being argumentative, just curious. In my personal view this is a good thing. I was in the early access to DALLE and without fail any searches for CEO or executive would return exclusively white people. Searches for low skilled jobs would return minorities. Surely such an obviously biased AI that is only going to get more popular should be steered away from this?
- niknoble 4y agoIf this is the behavior you want, you were always free to add these keywords to your prompts. OpenAI also could have made them optional as a checkbox. By appending them to every prompt, OpenAI strictly reduces what can be done with the tool. I would make that same complaint about any forcible changes to prompts, but it is actually even worse in the case. It is a concession to the ruling political regime of the United States, which will be applied worldwide, even to users who don't share our politics. Imagine if China had made the first great text-to-image generator, but it would automatically append keywords like "heterosexual" in order to prevent images that clash with Chinese culture. I don't think Americans would be too happy about that.