Hacker News Reader

Comment by qwertox | Hacker News Reader

Imagine the benefits a safe image generator would have on children, them learning to explain what they want to see, how it should be changed. A house with a tree and a dog and a lake nearby, instead of playing candy crush.

Why shouldn't it be solved? We clearly differ.

View on HN

If the children manage to induce LLM into generating NSFW content, they are probably capable of inducing their set of coloring pen into drawing dick shapes.

Now think about whether this is entirely the child’s idea or a suggestion or Rick roll by another kid or some stranger on the internet. If people are deploying these tools widely there will be more and more opportunities for that to happen.

I think it's not desirable to spend time fixing these elaborate scenarios. The same sort of kids trolling kids will happen via other means. At the end of the day, if a kiddo sees some porn, it's not the end of the world. Plenty of kids hear their parents having sex or will open a door they shouldn't have and see people in action. Making everything completely aseptic is only pandering to prude adults' discomfort.

And ultimately these LLM are tools. People are responsible for what they do with the tool. My Dewalt drill shouldn't try to decide which holes to cut, it's up to me. If I drill holes in the pattern of a dick shape, it's my fault. We don't hold back drill development because of their potential ability to create obscene art or depict gruesome situations.

Your Dewalt drill has a UL verification for having things like fuses and other design features which mean it won’t unexpectedly catch fire, too. That’s more what we’re talking about here: less “can a talented person make naughty art in photoshop?” than “if I integrate this into my educational tool or game, will I make the morning news because someone figured out how to get it to porn-roll the class or emit instructions for killing yourself?” or “will I regret allowing this site in my school’s firewall?”

These aren’t elaborate scenarios but rather the kind of things which happen constantly, and have affected product design for generations. Companies are hoping that this class of tool might be enough better to avoid those limitations, which is why this kind of research is important.

I see your point. It nuances my prior beliefs, although in the case being publicized, the LLMs spit out NSFW content when prompted elaborately. Kid friendly apps already have problems moderating chat and other user generated content, and Nintendo handles this by limiting the inputs to a set selection. Personally I think this is a marketing based fear more than a fear of real consequential damage being done to any kid. The businesses don't want to have their image associated with drama.

At least for the time being, I’ll put my trust in Crayola for that, rather than OpenAI. Alternatively, if we really must have children interacting with image generation, it seems like a great opportunity for parents to have some bonding time with their kids.

Reply on news.ycombinator.com