FilterHN

points

1 year ago

| 1 comment

This is an interesting point because it shows top-level control of output - by allowing one type of infringement rather than all types of infringement, it’s a human based decision. It’s only a matter of time before AI crosses into trademark domain issues and that’s where the big money lawyers will likely step in. As in, it harms the mark and it can be financially damaging unless stopped - and these systems clearly can be stopped from certain outputs.

▲

Maxatar

1 year ago

[-]

Don't immediately believe what you read on Hacker News, certainly not to the point where you'd jump to such a significant conclusion over it.

ChatGPT is simply inconsistent about enforcing rules. Sometimes you ask it to generate an image and it says no, and then you ask it again in a slightly different way and it happily does it for you. One time I asked it to generate an image for me and it told me "I can't generate an image of Elon Musk but I can generate an image of someone who looks like Elon Musk. Let me know if you'd like me to do that instead", and the generated image looked exactly like Elon Musk.