mesaThe AI feedback loop: Researchers warn of ‘model collapse’ as AI trains on AI-generated content
Collapse faster, please. Sick of ai bullshit clogging up my searches.
The Internet is fucked now, the only valuable untainted training data is the Internet as it existed prior to this AI bullshit coming online. Confirmed human content is going to be super valuable, so expect our privacy to be fucked as well…
Did anyone take a copy?
IninewCrowEven that is going to turn into a shit show … It will become a copy of a copy of a copy of a backup of a backup of a copy and all of it will just get rendered down to some common basics based on whatever the hell was marketed and promoted by bots
The collapse won’t stop ai output fron spamming the internet though. It will just make it worse and more lilely to he incorrect
We can just stop checking the internet for things
Really though, is does anyone know if/how Wikipedia is protecting itself from this?
It’s not going to. It’s just going to get more widespread and harder to detect. The incentives favor developing better and better AI. Luckily one of the solutions to this issue is - wait for it - AI. With a good enough AI, especially a generally intelligent one you don’t need search engines anymore. You just ask and it gives you the answer. If you think AI couldn’t do this reliably then that is not the AI I’m talking about.
Maybe we need to label AI-generated content to, you know, avoid confusion.
_haha_oh_wow_Sorry, best we can do is a race to the bottom fueled by greed and incompetence.
That will be a refeshing change.
That’s what has been happening, and is likely what will continue to happen. Not much change there really…
Yes that’s the joke.
Wouldn’t it have to be funny to be a joke?
I thought I was being funny. Sorry if it didn’t tickle you just right.
Please respect my personal space and refrain from tickling me.
Comedy is hard.
I'm sure we can compromise on a mandatory database of registered AI-generated content that only the corporations can read from but everyone using AI-generated content is required by law to write to, with hefty fines (but only for regular people).
Sounds great, how do we enforce it?
If the AIs want to avoid digital incest they’ll enforce it for themselves.
The AIs dont want anything themselves and those who make the decisions about them want the most profit, what costs more, verifying training data or AI incest?
tinwhiskersAh, you’re suggesting using RFC 3514. Good thinking.
Thank you for bringing that standard to my attention.
Oh goody. I’ve been wanting to use this since my slashdot days… today is my first chance!
Your post advocates a [x] technical [ ] legislative [ ] market-based [ ] vigilante approach to fighting (ML-generated) spam. Your idea will not work. Here is why it won't work. [One or more of the following may apply to your particular idea, and it may have other flaws which used to vary from state to state before a bad federal law was passed.] [ ] Spammers can easily use it to harvest email addresses [ ] Mailing lists and other legitimate email uses would be affected [ ] No one will be able to find the guy or collect the money [ ] It is defenseless against brute force attacks [ ] It will stop spam for two weeks and then we'll be stuck with it [ ] Users of email will not put up with it [x] Microsoft will not put up with it [ ] The police will not put up with it [x] Requires too much cooperation from spammers [x] Requires immediate total cooperation from everybody at once [ ] Many email users cannot afford to lose business or alienate potential employers [ ] Spammers don't care about invalid addresses in their lists [ ] Anyone could anonymously destroy anyone else's career or business Specifically, your plan fails to account for [ ] Laws expressly prohibiting it [x] Lack of centrally controlling authority for email^W ML algorithms [ ] Open relays in foreign countries [ ] Ease of searching tiny alphanumeric address space of all email addresses [x] Asshats [ ] Jurisdictional problems [ ] Unpopularity of weird new taxes [ ] Public reluctance to accept weird new forms of money [ ] Huge existing software investment in SMTP [ ] Susceptibility of protocols other than SMTP to attack [ ] Willingness of users to install OS patches received by email [ ] Armies of worm riddled broadband-connected Windows boxes [x] Eternal arms race involved in all filtering approaches [x] Extreme profitability of spam [ ] Joe jobs and/or identity theft [ ] Technically illiterate politicians [ ] Extreme stupidity on the part of people who do business with spammers [x] Dishonesty on the part of spammers themselves [ ] Bandwidth costs that are unaffected by client filtering [x] Outlook and the following philosophical objections may also apply: [ ] Ideas similar to yours are easy to come up with, yet none have ever been shown practical [ ] Any scheme based on opt-out is unacceptable [ ] SMTP headers should not be the subject of legislation [ ] Blacklists suck [ ] Whitelists suck [ ] We should be able to talk about Viagra without being censored [ ] Countermeasures should not involve wire fraud or credit card fraud [ ] Countermeasures should not involve sabotage of public networks [ ] Countermeasures must work if phased in gradually [ ] Sending email should be free [x] Why should we have to trust you and your servers? [ ] Incompatiblity with open source or open source licenses [x] Feel-good measures do nothing to solve the problem [ ] Temporary/one-time email addresses are cumbersome [ ] I don't want the government reading my email [ ] Killing them that way is not slow and painful enough Furthermore, this is what I think about you: [x] Sorry dude, but I don't think it would work. [ ] This is a stupid idea, and you're a stupid person for suggesting it. [ ] Nice try, assh0le! I'm going to find out where you live and burn your house down! ````___`I traced this baby back to March 2nd, 2004: craphound.com/spamsolutions.txt
Thanks, I was wondering how old it was when they said “Slashdot days.”
Oh, do me next, do me. Open source adversarial models trained to detect and actively label things which it detects as belonging to AI. Probably would end up looking like a browser extension or something. Ublock, but for AI, basically.
Far too late for that now.
Can you explain further more ? Sorry I may understand but not sure
Low Background Radiation Steel was/is valuable, because it’s made of steel from before nuclear testing. As the bombs contaminated the produced steel.
In the same sense, anything before the creation of LLMs would be considered “low background radiation” content, as that’s the only content to be sure to be made without LLMs in the loop
I like the concept and analogy, thanks for explanations 🙏
I can just take a picture with my phone whenever I want. This is a bad analogy
DojanI mean it makes sense. Machine learning is fantastic at noticing patterns, and the stuff they generate most definitely do have patterns. We might not notice them, but the models will pick up on them and eventually, if you keep training them on that data, they’ll skew more and more in that direction.
They’ve been marketing things like there isn’t a limit to how good these things can get, but there is. Nothing is infinite.
circuitfarmerI’ve tried to make this point several times to folks in the industry. I work in AI, and yet every time I approach some people with “you know it ultimately just repeats patterns”, I’m met with scoffs and those people telling me I’m just not “seeing the big picture”.
But I am, and the truth is that there are limits. This tech is not the digital singularity the marketers and business goons want everyone to think it is.
It can be both though.
It repeats things that sort of sound intelligent to try and convince everyone that actual intelligent thought is taking place? It really is just like humans!
They don’t really parrot unless they’re overfitted.
It’s more that they have been trained to produce a certain kind of result. One method you can train them on is by basically assigning a score on how good the output is. Doing this manually takes a lot of time (Google has been doing this for years via captcha), or you could train other models to score text for you.
The obvious problem with the latter solution is that then you need to ensure that that model is scoring roughly in line with how humans would score it; the technical term for this is alignment. There’s a pretty funny story about that with GPT-2, presented in a really cute animation format by Robert Miles.
The True Story of How GPT-2 Became Maximally Lewd
Tell me about it. All the government contractors I work with. Just repeating the same submittal over and over and over again.
Legend says that humans developed pattern finding as a skill ages ago…
FaceDeerThis article is from June 12, 2023. That's practically stone-aged as far as AI technology has been progressing.
The paper it's based on used a very simplistic approach, training AIs purely on the outputs of its previous "generation." Turns out that's not a realistic real-world scenario, though. In reality AIs can be trained on a mixture of human-generated and AI-generated content and it can actually turn out better than training on human-generated content alone. AI-generated content can be curated and custom-made to be better suited to training, and the human-generated stuff adds back in the edge cases that might disappear when doing repeated training generations.
That’s it! I’m starting my own internet, with blackjack. And hookers.
I fail to see how that differs from the current Internet
It’s more Futurama!
that’s why the fediverse exists
Back when i was though concept art as a subject at college my teacher had a name for this.
“Incest” cause every generation of art that references other art becomes more and more strange looking and detached from reality.
If you thought Skyrim weapons look ridiculous you should have seen my classmates Skyrim inspired weapons.
If you think that looked ridiculous, you should have seen the Skyrim weapons inspired by your classmates weapons.
My team has been calling models that use ai generated data “Habsberg models”
Lmao that’s a perfect name for it
I feel there is a good joke here, but I miss the knowledge to understand it. Care to enlighten me?
The Habsburg royal family line was famously inbred with a distinctive chin.
vaultdweller013For anyone curious on how bad it got look up the coroners report on Charles of Spain. Its fucken grizzly.
Anecdotally speaking, I’ve been suspecting this was happening already with code related AI as I’ve been noticing a pretty steep decline in code quality of the code suggestions various AI tools have been providing.
Some of these tools, like GitHub’s AI product, are trained on their own code repositories. As more and more developers use AI to help generate code and especially as more novice level developers rely on AI to help learn new technologies, more of that AI generated code is getting added to the repos (in theory) that are used to train the AI. Not that all AI code is garbage, but there’s enough that is garbage in my experience, that I suspect it’s going to be a garbage in, garbage out affair sans human correction/oversight. Currently, as far as I can tell, these tools aren’t really using much in the way of good metrics to rate whether the code they are training on is quality or not, nor whether it actually even works or not.
More and more often I’m getting ungrounded output (the new term for hallucinations) when it comes to code, rather than the actual helpful and relevant stuff that had me so excited when I first started using these products. And I worry that it’s going to get worse. I hope not, of course, but it is a little concerning when the AI tools are more consistently providing useless / broken suggestions.
There will soon be a filter on the “best” developers, if there isn’t one already.
Politically IncorrectThe “AI” armageddon…
If the AI generated content is labeled, or has context, or has comments or descriptions created by people, then wouldn’t it just be the same as synthetic training data? Which is shown to still be very useful for training.
Most AI-generated data in the wild won’t have labels because there’s no incentive to label it, and in a lot of cases there are incentives to not label it.
The “solutions” to model collapse - essentially retraining on the original data set - suggests LLMs plateau or deteriorate. Especially without a way to separate out good and bad quality data (or ad they euohemistically try and say human vs AI data).
Were increasingly seeing the limitations and flaws with LLMs. “Hallucinations” or better described as serious errors, model collapse and complete collapse suggest the current approach to LLMs is probably not going to lead to some gone of general AI. We have models we don’t really understand that have fundamental flaws and limitations.
Unsurprising that they probably can’t live up to the hype.