[jerf]

This is the fundamental reason why I am in favor of a ban on simply posting AI-generated content in user forums. It isn't that AI is fundamentally bad per se, and to the extent that it is problematic now, that badness may well be a temporary situation. It's because there's not a lot of utility in you as a human being basically just being an intermediary to what some AI says today. Anyone who wants that can go get it themselves, in an interactive session where they can explore the answer themselves, with the most up-to-date models. It's fundamentally no different than pasting in the top 10 Google results for a search with no further commentary; if you're going to go that route just give a letmegooglethat.com link. It's exactly as helpful, and in its own way kind of carries the same sort of snarkiness with it... "oh, are you too stupid to AI? let me help you with that".

Similarly, I remember there was a lot of frothy startup ideas around using AI to do very similar things. The canonical one I remember is "using AI to generate commit messages". But I don't want your AI commit messages... again, not because AI is just Platonically bad or something, but because if I want an AI summary of your commit, I'd rather do it in two years when I actually need the summary, and then use a 2027 AI to do it rather than a 2025 AI. There's little to no utility to basically caching an AI response and freezing it for me. I don't need help with that.

[code_biologist]

It's been interesting to watch this play out in microcosm in different spaces. Danbooru and Gelbooru are two anime image boards that banned AI image content, largely to their benefit in my opinion. Rule34 is a similar image board that has allowed AI images and they've need to make tagging and searching adaptations to add to handle the high volume of AI images versus human artists. I'm glad there's an ecosystem of different options, but I find myself gravitating to the ones that have banned AI content.

[verall]

I fully agree with this, besides that if an AI could auto-generate a commit message that I can edit to make actually correct and comprehensive, it will probably be a better, more descriptive message than whatever I come up with in usually ~3 minutes.

The value is a nice starting point but the message is still confirmed by the actual expert. If it's fully auto-generated or I start "accepting" everything, then I agree it becomes completely useless.

[9rx]

> It's because there's not a lot of utility in you as a human being basically just being an intermediary to what some AI says today.

To be fair, there has never been a lot of utility in you as a human being involved, theoretically speaking. The users do not use a forum because you, a human, are pulling knobs and turning levers somewhere behind a meaningless digital profile. Any human involvement that has been required for the software to function is merely an implementation detail. The harsh reality, as software developers continually need to be reminded of, is that users really don't care about how the software works under the hood!

For today, a human posting AI-generated content to a forum is still providing all the other necessary functions required, like curation and moderation. That is just as important and the content itself, but something AI is still not very good at. A low-value poster may not put much care into that, granted, but "slop" would be dealt with the same way regardless of whether it was generated by AI or hand written by a person. The source of content is ultimately immaterial.

Once AI gets good, we'll all jump to AI-driven forums anyway, so those who embrace it now will be more likely to stave off the Digg/Slashdot future.

[briangriffinfan]

I notice this idea repeatedly popping up, that "content made by humans" is more-or-less equivalent to content generated by genAI inasmuch as genAI can "in theory" (in some sense) create the same thing a human could and so they're both "in theory" the same thing.

The idea that words people write don't mean anything or imply anything in an abstract sense is misguided, in my opinion. When one reads something a person wrote, they think about what the person who wrote it was thinking, what it means to them, what the implications of what they think might be... there are people who do not think about things like this, so they don't care and view genAI text as equivalent because that level of thought simply isn't put into their reading.

Anyway, my point is talking on a forum filled with LLMs would probably stop being interesting and engaging very quickly because LLMs are bad at emulating the lateral thinking, diversity of ideas, and abstraction of communication that make talking to a human fun.

[9rx]

> because LLMs are bad at emulating the lateral thinking, diversity of ideas, and abstraction of communication that make talking to a human fun.

While that is no doubt true today, the earlier comment that sparked this posits that may only be a temporary state that may improve in the future. Once LLM and human creation is indistinguishable, there is no reason to have concern for what generated the content, is there?

Nobody uses a forum for the human connection. There is no human! I can't see your face, I can't touch your skin, I don't feel the heat radiating from your body. Hell, if we meet each other on the street later today, I'll never know it was you. There is only software. I do assume, knowing a thing or two about how the technology works, that in implementation that there is a human somewhere in the loop, but I don't completely know for sure, and it wouldn't make a difference anyway.

There is a place for human connection, most certainly, but it is found in the "real world". Forums are not equivalent. They are something else entirely.

> Anyway, my point is talking on a forum filled with LLMs would probably stop being interesting and engaging very quickly

Just as it does when humans write drivel by hand. There is merit to banning accounts that post garage, but what produced that garbage is irrelevant. AI's involvement, or lack thereof, makes no difference. The quality of an account can be judged on its output, not the mechanism by which it operates.

[briangriffinfan]

> Just as it does when humans write drivel by hand.

I see this a lot in AI discussions - it can, at best, do what we do at a level we'd consider "good enough." It can write mediocre slop just as well as the most mediocre of us. To me, that is lowering the bar for our work exceptionally. We shoot for the stars; we just miss sometimes!

[9rx]

You don't have to lower the bar in order to accept AI, but where AI is able to meet the bar, there is no logical reason to ban it. The user really doesn't care about how the software is implemented.

I understand the luddites are always fearful of losing their job, but that is ultimately irrational.