David L 💙💛

Software engineer, mild Hearthstone addict, public account for @dmclap. Opinions are my own, as far as you know. He/him @dmclap.bsky.social

[Shaun 🏴󠁧󠁢󠁳󠁣󠁴󠁿🇺🇸 @shaunvlog_]

Every Scottish person in America needs to immediately try Chicken Fried Steak, and you’ll realise we and the Americans are kindred spirits

[Nav Toor @heynavtoor]

You have noticed it. ChatGPT feels dumber than it used to. Your prompts that worked six months ago produce worse results now. The writing sounds flatter. The ideas sound safer. The internet itself feels like it is shrinking. Every article reads the same. Every email sounds the same. Every answer sounds like it was written by the same voice.

You thought it was you. It is not you.

Researchers at Oxford and Cambridge published a paper in Nature proving what is happening. They call it Model Collapse.

Here is the mechanism in one sentence. AI trained on AI-generated data gets dumber every generation until it forgets what real human data looked like.

The internet is filling with AI-generated content. Blog posts. Articles. Reviews. Comments. Social media. AI companies scrape the internet to train the next generation of models. Which means the next generation of AI is being trained on the output of the current generation.

Each cycle loses information. Not randomly. It loses the rarest, most unusual, most creative parts first. The researchers call these the "tails of the distribution." The weird ideas. The unexpected perspectives. The things that made the internet feel human. Those disappear first.

What remains is the average. The safe. The expected. The bland.

Then the next generation trains on that. And loses more. And the next generation trains on that. And loses more. The researchers proved this is not a slow decline. Major degradation happens within just a few iterations. Even when some of the original human data is preserved.

They tested it on large language models. On image generators. On statistical models. The pattern was the same every time. The output converges toward a narrow, flattened version of reality that looks nothing like the original data.

The lead researcher put it plainly. "Large language models are like fire. A useful tool. But one that pollutes the environment."

The pollution is invisible. You cannot see which sentence on the internet was written by a human and which was written by AI. Neither can the AI that is about to train on it. And once the tails are gone, they do not come back. The damage is irreversible.

This is not a prediction anymore. It is a diagnosis.

The internet you grew up on was built by humans writing things no algorithm would have written. Strange, personal, imperfect, alive. That internet is being diluted. One generation of AI at a time. And the models trained on what remains are learning a smaller and smaller version of the world.

Model Collapse is not a technical problem. It is a cultural one. The thing that made the internet worth reading is the thing that disappears first.

[Gayest Tone @gayest_tone]

There’s basically two stories about the World Cup right now, it’s either

“US Customs hooks Tajik team up to the BallGrabber 7000 as they exit bus”

or

“59 year old man in Bentonville excitedly invites entire Congolese team to have breakfast with him at Waffle House”

[one dozen rats at a keyboard @PanasonicDX4500]

there’s a kernel of truth here in that most Union soldiers didn’t really care about slavery until they got down to the South and saw it firsthand, at which point they had to be regularly stopped from lynching the Plantation owners

[John Scott-Railton @jsrailton]

NEW: malware developers added nuclear & biological weapons text to to their spyware.

Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner.

Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky.

When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit.

We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted.

In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation.

H/T to colleagues that shared this with me https://t.co/f3Aj9TYxU4

[Corey Quinn @QuinnyPig]

The funniest possible outcome of the AI mandate era is about to be HR departments discovering that "sincerely held religious belief" under Title VII has a much lower bar than they assumed, and Pope Leo handed every Catholic employee a written excuse.

[David L 💙💛 @GroundRuleSEV2]

@DivineWolfwood Yeah, all totally fair, and I definitely don't discount your experience. I'm just know that United, American, and Alaska have all switched to group numbers without mentioning rows, and those are the main ones I fly. Which is why I was curious

[David L 💙💛 @GroundRuleSEV2]

@DivineWolfwood Ok yeah, I figured that might be the case. Which airline is it? Because the other alternative is that the back is the Economy Peasant class or something that doesn't get to use the overhead bins (and possibly not the bathrooms)

[David L 💙💛 @GroundRuleSEV2]

@DivineWolfwood Now that you mention it, are you sure it's not back to front? I feel like all domestic flights board by nebulous Zones now, which is more about ticket class an elite status than anything else. And I feel like my last international flight went back to front

[David L 💙💛 @GroundRuleSEV2]

@DivineWolfwood Because they want the people in first class to implicitly stunt on us hoes in the back. Also because people are morons and they can't use a more sophisticated system

[Tim Miller @Timodc]

Put some respect on Schumer/Jeffries name. The only reason the Senate has leverage to fight the Slush Fund & Ballroom is because ICE/DHS wasn't funded. How does the line go? Let the apology be as loud as the disrespect?

More from @EggerDC

https://t.co/DGrxSt4T3g

[Josh Barro @jbarro]

"Powered by Polymarket" bug on this incendiary lie (no indication the shooters were trans and their manifesto says "for the trannies, you disgust me") -- worth considering whether you consider Polymarket a reputable brand before doing business with them.

[Jeremiah Johnson 🌐 @JeremiahDJohns]

Public sector unions are bad.

These unions are not negotiating against billionaires or some convenient cartoon villain. They're rent seeking and negotiating against taxpayers, and they'll shut down vital infrastructure if we don't humor their tantrum.

[David L 💙💛 @GroundRuleSEV2]

@Pinboard If they're trying to acclimate US soldiers to their greatest threat, they should have truck salesmen with 28% interest loans roaming the training grounds

[Andy Masley @AndyMasley]

This one's a completely real problem, the data center is technically under the noise threshold where the county acts but produces enough constant annoying low level hum that it's lowering quality of life for the homes nearby. A lot of the noise is coming from temporary gas turbines that will be gone once it's fully connected to the grid, but that timeline's been extended way back and could be as much as 7 years now. This is a ridiculous situation that imo a lot of places don't have good rules to govern well right now.

[David L 💙💛 @GroundRuleSEV2]

@FischerKing64 You took a dumb stance and made it dumber. Incredible

[one dozen rats at a keyboard @PanasonicDX4500]

If I have to make peace with my neighbor having an uninfringeable right to an AR-15 and a dozen semi-automatic handguns, you’ve gotta make peace with your neighbor being a guy with Honduran parents who was born in New Jersey, sorry.

[ProPublica @propublica]

The Flex Loan, a new type of payday loan pioneered by Advance Financial in Tennessee, allows residents to borrow up to $4,000 at a 279.5% interest rate.

It has burdened low-income borrowers while generating huge profits for lenders.
https://t.co/mVCStJqu8n