Donald Trump discovers diffusion.

It’s like I’m living in a zoo…

This is hilarious. It’s like the Iranians using (and paying for) Windows OneDrive for disinfo efforts, when they could use literally anything else.

Don’t they actually have competent hackers though? I guess the knowledge intersection between the hacking space and LLM land is less than I thought.

They have always hosted AVC videos. They have to for old/incompatible devices, because literally any toaster and software on earth will play it.

But sometimes they are funny about what gets encoded to what, and of course they will always try to default to av1/vp9 and opus.

Surely it would fall back to mp4? Are some videos AV1/VP9 only?

I dunno if I would go that far, but Bibi all but sabotaging Biden does look bad for Biden. Biden must be fuming over the assassination.

Yeah, even if it ends tomorrow (which it will not, not even close) this has totally screwed over Russia.

So they’d become a kind of vassal state?

I could totally see that, especially if something happens to Putin.

My hot take:

Whatever benefits Bibi is usually what happens. As an outsider, it seems like conflict keeps him in office, so something will coincidentally happen to make this fall through… probably.

They already couldn’t afford this situation, and look where they are.

What’s an improbable “acceptable risk” to them may not be good enough for NASA, especially if they don’t really understand what’s wrong.

Musk is a Musk guy.

That narrative benefited him before (because Tesla. Maybe SpaceX in a really weird way). Now it doesn’t.

I’ve driven by Mar-a-Lago, and they key it sits on is almost like a cliff. He’s also on the “coastal” side of the key which protects it from the open ocean a little.

It’s a ridiculously wealthy area too, so they’ll just fix hurricane damage and erosion. TBH it’s not something he has to worry about there.

A very nuclear power at that.

WTF, what was the justification for that rule before they changed it?

This doesn’t touch the weights at all, it’s just a change to the sampler.

What lobotomizes their models is cost cutting and trying to make them “safe,” or at least thats what I suspect.

They can cycle a some biases (dozens?) and test them all. Detokenization is super cheap to run, its not AI or anything.

I’m trying to think of a good analogy for how this would work, and I kinda came up with one. This would be kinda like an image encoder that biases itself towards coding RGB values (0-255) as even numbers. Subtly, say 30% odd 70% even.

That’s totally imperceptile to humans. And even a “small” sample of the image would carry this bias if pasted into a larger image verbatim, since the sample size is so large (just as the sample size for a bunch of tokens in text is pretty big.

And I’m not saying its fullproof… but if thats indeed what they’re doing, I think its a decent way to detect “lazy” OpenAI abusers who aren’t working so hard to scramble and defeat it.

It’s not so trivial if OpenAI cycles the logit bias or makes it really convoluted.

And it’s not like certain “words” or language patterns are more probable with this method, its different than what any kind of human or words based algorithm would detect, which is what I suspect most “anti AI detection” software does.

Its doable… but seems inconvenient for a small business to keep up with. Maybe.

You have full control of your logit outputs with local LLMs, so theoretically you could “unscramble” them. And any finetuning would just blow that bias away anyway.

OpenAI (IIRC) very notably stopped giving the logprobs of their models. They did this for many reasons, and most of them boil down to “profits” and “they are anticompetitive jerks,” but another reason is to enable watermark methods just like this.

Also, thing about this is that basically no one uses self hosted LLMs compared to OpenAI (or really any API) LLM.

This has been known in the ML space forever. LLMs don’t actually output words/tokens, but probabilities for a long list of tokens, and the sampler picks one (usually the mostl likely token). And if you arbitrarily weigh these probabilities (eg 50% of possible token outputs are more likely than the other 50%, as a random example), it creates a “signature” in any text thats easy to measure. The sampler randomizes it a tiny bit, but that averages out in long texts.

It’s defeatable. I’m sure if you maken enough OpenAI queries, you can find the bias. I think a paper already tackled this. But this likely will stop the lazy absures, aka 99% of abusers, who should just use some other LLM if they really care.

Another open secret in LLM land is that OpenAI is actually falling behind open research efforts, hence its hilarious it took them this long to implement something so simple.

Yeah, well… Bibi doesn’t. Not unless it helps him stay in office.

Lemmy is so sour on this idea, presumably because it was Trump’s first.

It seems fine? Like others said, its hard to enforce anyway.