As a reminder, these things run on normal computers, and they are giving them away for free.

Does your company have computers, y/n?

No shit, at this point. They’re inscrutable probabilistic chatbots and the “guardrails” are mostly in-band communication.

And this is for tactics that occasionally work against humans. ‘We can neither confirm nor deny there was a warhead.’ How big was this warhead? ‘Three megatons.’ Big surprise you can outsmart a program which people insist is not any form of intelligent.

This is why feeding in public data is fine - like, if Googling hard enough could turn up Geocities instructions for making meth, then that’s not really a secret - but any use of private information is a data breach with more steps.

An improvement, but fuck off nonetheless.

Let’s be honest, things could’ve gone better.

Conspiracy wank.

Gotta wonder where low-framerate rendering and FLF generation could outpace rendering alone. Probably a less impressive use-case thanks to Eevee.

Furthermore he declares base, and no whammies, and these shallow nuh-uhs will totally be legally binding as soon as his fat ass is gone.

This is approaching unfalsifiable.

Is classifying generated data ever intractable, when you can just… generate some? Even if the portion of human content dwindled - rather new human content, since there’s a recent hard cutoff - there’s a faucet for counterexamples. You can have as much for-sure machine-generated content as you like, to point at and go, wrong.

… and identification criteria of ‘yeah he looks similar.’

I again submit the last two years where model collapse did not happen. The doom-and-gloom predictions - some rather gleeful - plainly missed the mark. The proliferation of generated content has not in fact ruined the content generators, and it’s sure not because we’re any good at marking generated content. Early symptoms went away entirely and the problem has been practically addressed.

As for “unlearning,” universality is why it’s a made-up problem. Nobody loudly complains that x-rays make doctors worse at feeling around for lumps.

That’s a lot of “could” and “will” from an article a year old, primarily about concerns from two years ago, while image models to-day keep getting smaller and better. They didn’t find a second internet’s worth of JPEGs. Better training on the same data, or even better labels on less data, beats a simple obsession with scale.

Yes, photocopying a photocopy will degrade, but diffusion is a denoising algorithm. Un-degrading an image is its central function. ‘Make it look less AI’ is how you get generative adversarial networks.

Anyway, the grim truth is that the central concern is mistaken. Training data for cancer screening does not require the patient lived.

The doctors were better, until someone yanked the tool away. That’s how every tool works! Even going from a handsaw to a table saw and back will make you lose some skill with the handsaw, because your brain focused on higher-level goals and finer motions. That’s not proof a table saw is bad for woodworking. The problem is “and back.”

since apparently AI can’t feed into AI without collapse

Have you checked on that narrative? It’s been a while. Things stopped getting yellow. Improvements continued.

Take away a calculator halfway through an exam, and suddenly, people are surly and unmotivated about simple long division.

That’s how every ‘AI makes you stupid!’ article works. Like, ‘doctors used AI to detect more cancer, but when we took it away, they were worse at eyeballing it.’ Sorry, can we go back to the part about detecting cancer better?

‘Well if it’s merely okay just have it be perfect.’

Unserious.

The argument is, fuck Nvidia’s monopoly on GPU compute. AMD keeps quietly funding projects to simply translate CUDA, then backing off when that funding is discovered. That’s how scared they are of a court battle. And that was before the AI bubble’s infinite money glitch.

Now even more of your own code can become proprietary and single-vendor!

Reminded of I Am Legend’s premiere, where $6.87 gas got a nervous laugh from the audience.

And then years later we actually hit that price through normal economic trends, and chuds blamed Biden like it was his fault.

And now those same chuds refuse to blame The Idiot for surpassing that price as unambiguous consequences of his warmongering.

I would have preferred the zombies. Even if they do that late-2000s CGI stretchy jaw thing.

James May tried!

Tools can be useful well before they’re taken for granted. Art software was always a hot mess. 3D software’s still a hot mess. People nonetheless find immense utility in these programs.

This tech would be a non-issue if it didn’t actually work. Chatbots can code now, and they’re good enough that I’ve seen critics fixate on maintainability, which is about as high-level as complaints could be. The big fat datacenter versions have caused sharp divisions by reimplementing open-source projects, using completely different structure in other languages entirely. The offline laptop versions are only months behind. Shit is getting weird in this house.

I was taught with compilers you couldn’t trust. The old buggy ones that needed babysitting still beat doing everything by hand.

Even when they worked, you were expected to know assembly for inline optimizations, and my assembly education was surely terrible compared to my teachers’. That gap was okay, because things usually worked out. The point was using the higher-level tools to get more done without sweating every last detail.