FaceDeer

You didn't read the article. Microsoft is not going to be running the reactor.

Until people stop mindlessly clicking on bait about Microsoft, nuclear power, and AI being bad.

The implicit guardrails these companies are going to add which will complicate things.

That'll just have to be part of evaluating whether a game is "good" or not, I guess. If game companies hobble their NPCs with all sorts of limitations on what they can talk about then it'll harm the reception of the game and drop its metacritic score.

I do see some interesting hurdles that were likely never imagined when the rules were written. How do you come up with an ESRB rating for a game where you don't know what topics your NPCs might talk about or what sorts of quest lines might ultimately be generated?

Numerous game-breaking states because you're risking a more traditional Dungeons & Dragons Dungeon Master problem where your party somehow has failed to ask an NPC the right kind of questions or even consider that they might have information relevant to the campaign. How do you get this information across if the player isn't somehow prompted to attempt it?

That seems like something that an AI-driven game might actually be better at, if properly done. The AI could review the dialogue the character has participated in so far and ask itself "has the player found out the location of the cave with Necklace of Frinn yet?" And if it sees that the player just keeps on missing that vital clue somehow it could start coming up with new ways to slip that information into future dialogues. Drop hints and clues, maybe even invent a letter to have delivered to the player, that sort of thing.

Whereas in a pre-scripted game if a player misses a vital clue they might end up frustrated and stuck, not knowing they need to backtrack to find what they overlooked.

I think this AI stuff is a cheap cop-out that uses way too much energy for a weak result.

If the games using AI aren't good then they won't sell well. This is a self-correcting problem.

We have, but of course not to the satisfaction of anti-nuclear activists because solving it would be counter to their actual goals.

Nuclear waste is actually quite easy to deal with unless your purposes are best served by it being very difficult to deal with, in which case you make as much trouble as you can for it.

Also, what do you mean by synthetic data? If it's made by AI, that's how collapse happens.

But that's exactly my point. Synthetic data is made by AI, but it doesn't cause collapse. The people who keep repeating this "AI fed on AI inevitably dies!" Headline are ignorant of the way this is actually working, of the details that actually matter when it comes to what causes model collapse.

If people want to oppose AI and wish for its downfall, fine, that's their opinion. But they should do so based on actual real data, not an imaginary story they pass around among themselves. Model collapse isn't a real threat to the continuing development of AI. At worst, it's just another checkbox that AI trainers need to check off on their "am I ready to start this training run?" Checklist, alongside "have I paid my electricity bill?"

The problem with curated data is that you have to, well, curate it, and that's hard to do at scale.

It was, before we had AI. Turns out that that's another aspect of synthetic data creation that can be greatly assisted by automation.

For example, the Nemotron-4 AI family that NVIDIA released a few months back is specifically intended for creating synthetic data for LLM training. It consists of two LLMs, Nemotron-4 Instruct (which generates the training data) and Nemotron-4 Reward (which curates it). It's not a fully automated process yet but the requirement for human labor is drastically reduced.

the only way to guarantee training data isn't from its own model is to make it yourself

But that guarantee isn't needed. AI-generated data isn't a magical poison pill that kills anything that tries to train on it. Bad data is bad, of course, but that's true whether it's AI-generated or not. The same process of filtering good training data from bad training data can work on either.

It's not wrong for either to draw inspiration from the other. It's the hypocrisy that's wrong.

I've made similar points in the past in discussions about robot soldiers going to war. There's an upside to these things that people insist on overlooking; they follow their programming. If you program a robot soldier to never shoot at an ambulance, then it will never shoot at an ambulance even if it's having a really bad day. Same here, if the security robot has been programmed never to leave the public sidewalk then it'll never leave the public sidewalk.

It's always possible for these sorts of things to be programed to do the wrong things, of course. But at least now we have the ability to audit that sort of thing.

Are you suggesting that the same amount of crime is happening but they're deciding not to report it because there's a robot there? That's the measure they're touting, the reduction in crime reports.

You joke, but presumably that's when it recharges.

They're not both true, though. It's actually perfectly fine for a new dataset to contain AI generated content. Especially when it's mixed in with non-AI-generated content. It can even be better in some circumstances, that's what "synthetic data" is all about.

The various experiments demonstrating model collapse have to go out of their way to make it happen, by deliberately recycling model outputs over and over without using any of the methods that real-world AI trainers use to ensure that it doesn't happen. As I said, real-world AI trainers are actually quite knowledgeable about this stuff, model collapse isn't some surprising new development that they're helpless in the face of. It's just another factor to include in the criteria for curating training data sets. It's already a "solved" problem.

The reason these articles keep coming around is that there are a lot of people that don't want it to be a solved problem, and love clicking on headlines that say it isn't. I guess if it makes them feel better they can go ahead and keep doing that, but supposedly this is a technology community and I would expect there to be some interest in the underlying truth of the matter.

No, researchers in the field knew about this potential problem ages ago. It's easy enough to work around and prevent.

People who are just on the lookout for the latest "aha, AI bad!" Headline, on the other hand, discover this every couple of months.

AI already long ago stopped being trained on any old random stuff that came along off the web. Training data is carefully curated and processed these days. Much of it is synthetic, in fact.

These breathless articles about model collapse dooming AI are like discovering that the sun sets at night and declaring solar power to be doomed. The people working on this stuff know about it already and long ago worked around it.