Hacker Newsnew | past | comments | ask | show | jobs | submit | Legend2440's commentslogin

This is a toy example of how it could happen, in an artificial setting where you train entirely on generated outputs many times in a row.

It does not say that it is happening in production LLMs. It is a theoretical concern right now.


Where proof? Article contains zero evidence this is actually happening, just reasons he thinks it must happen.

Also I think this article itself may be AI-generated.


The fact that you aren't persuaded is the evidence of collapse. Previous generations of LLMs persuaded everyone of anything.

No, that's just hedonic adaptation of humans, not evidence of model collapse.

>Remember that the majority of industry is upstream of consumption.

People forget this. Oil companies may have dug up the oil, but they did so because we paid them to, so we could use the energy for good and useful things.

Climate change isn't 'evil billionaire companies are ruining the world', it's 'these things we did to improve our lives turn out to have side effects'.


The discussion is about the current generation of LLMs. It's not yet clear whether side-effects outweigh the advantages.

OTOH, I can already argue with numbers at hand that Bitcoin made the world poorer and worse off.


This is backwards. If it weren't for 'eliminating jobs' we'd both be peasant farmers right now. Automation has improved the standard of living and raised wages for everyone, rich and poor alike.

Ok, Claude.

I don't think you realize how bad NLP was prior to transformers. Oldschool entity recognition was extremely brittle to the point that it basically didn't work.

CV too for that matter, object recognition before deep learning required a white background and consistent angles. Remember this XKCD from only 2014? https://xkcd.com/1425/


CV is a space where I would 100% agree with you. But - edge cases notwithstanding - there's not so much of a dropoff with NER that I would first go to an LLM.

Ok, sure, it's usually cosmetic. So what?

Certainly it's not impossible to DIY, but it's more difficult than just popping some aligners on your 3d printer.

Manufacturing them requires a resin printer and a vacuforming setup, but that's still the easy part. It's a whole system with a dental 3D scanner, software for rearranging your mouth, and attachment points that have to be epoxied onto (and later removed from) your teeth by a dentist.


They have to have at least 2 different materials as well. The temporary trays were much softer and I had almost ground through them in my sleep by the time I had to switch to the next one but the final set is much more robust.

Yeah it's also not unreasonably expensive. At least when I had them it was only a few thousand pounds. I think they do offers regularly.

>It's also newsworthy when one of the most powerful and expensive weapon platforms in history doesn't have its infosec buttoned down.

Well, peace makes you sloppy. No one is at war with France right now, and no one is realistically going to attack this ship.

If we were fighting WW3, you can bet sailors wouldn't be allowed to carry personal cellphones at all. Back in WW2, even soldier's letters back home had to be approved by the censors.


Yawn. I'm tired of dystopian fiction. We're likely to get something that is neither dystopia nor utopia, but somewhere in between.

No, they have a decentralized crypto-based system where people propose a resolution to the bet, then token holders vote.

https://docs.polymarket.com/concepts/resolution


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: