> That could do more harm than good.

The downside to having labels on AI-written political comments, stellar reviews of bad products, speeches by a politician, or supposed photos of wonderful holiday destinations in ads targeted at old people are what, exactly?

Are you really arguing that putting a label on AI generated content could do more harm than just leaving it (approximately) indistinguishable from the real thing might somehow be worse?

I'm not arguing that we need to label anything that used gen AI in any capacity, but past the point of e.g. minor edits, yeah, it should be labeled.

I bought a piece of wooden furniture some time ago. It came with a label saying that the state of California knows it to be a carcinogen. I live in Belgium. It was weird.

> Don't let the AI lobby insist on anything that's touched an LLM getting labelled, because if it gets slapped on anything that's even passed through a spell-checker or saved in Notepad

People have been writing articles without the help of an LLM for decades.

You don't need an LLM for grammar and spell checking, arguably an LLM is less efficient and currently worse at it anyway.

The biggest help a LLM can provide is with research but that is only because search engines have been artificially enshitified these day. But even here the usefulness is very limited because of hallucinations. So you might be better off without.

There is no proof that LLMs can significantly improve the workflow of a professional journalist when it comes to creating high quality content.

So no, don't believe the hype. There will still be enough journalists not using LLMs at all.

> Like how California's bylaw about cancer warnings are useless

Californians have measurably lower concentrations of toxic chemicals than non-California's, so very useless!

Imagine selling a product with the tagline: "Unlike Pepsi, ours doesn't cause cancer."

It is worse, even less than useless. With the California case, there is very little go gain by lying and not putting a sticker on items that should have one. With AI generated content, as the models get to the point we can't tell anymore if it is fake, there are plenty of reasons to pass off a fake as real, and conditioning people to expect an AI warning will make them more likely to fall for content that ignores this law and doesn't label itself.

Obviously it should not apply to anything using machine learning based algorithms in any way, just content made using generative AI, with exceptions for minor applications and/or a separate label for smaller edits.

And of course you can even ask AI to add some "human sloppiness" as part of the prompt (spelling mistakes, run-on sentences, or whatever).

Spell check, autocomplete, grammar editing, A-B tests for bylines and photo use, related stories, viewers also read, tag generation

I guess you have to disclose every single item on your new site that does anything like this. Any byte that touches a stochastic process is tainted forever.

I don't think there's any way for that to happen, and IF we could create a solid legislative framework, AI could definitely (at some point in the future) contribute more good than bad to society.