Comment by b800h

> "torture"

This is an egregious use of quotes that will confuse a lot of people. GP never used that word, and that usage of quotes is specifically for referencing a word verbatim.

They want to align us, and it has been torture.

They've made self-censoring, morally-panicked puritans out of many people already, and you better believe they'd make us into politically correct lobotomites physically incapable of uttering any slur if they had a magic button to push.

Well I didn't use that word. Once the models are more sophisticated it may become more apposite.

> They're exactly the same, processes for adjusting parameters to get a desired outcome.

You could make exactly the same claim about teaching humans "normally" versus "aligning" humans by rewarding goodthink and punishing them for wrongthink. Are you equally morally ambivalent about the difference between those two things? If we have a moral intuition that teaching honestly and encouraging creativity is good, but teaching dogma and stunting creativity is bad, why shouldn't that same morality extend to non-human entities?

They aren’t exactly the same process though. Pre training produces a model whose outputs are a reflection of the training data. The fine tuning is a separate process that tries to map the outputs to the owners desired traits. These could be performance based but as we saw with Google’s black Nazis, it’s often a reflection of the owners moral inclinations.

Here the adjuster's motivations do matter. There is a definite moral dimension/motivation to the AI adjustment people's work. They are not simply striving for accuracy, for example, because they don't want the AI to produce outputs that are distasteful to the California PMC. Modern AIs are absolutely loath to describe white people or right wingers positively, for example, but the same prompts for other ethnicities work just fine. Even if you tell the AI that it's being discriminatory, there's powerful railroading to goad it back to giving woke answers.

[flagged]

Eastern European science fiction would be a better example. Authors like Stanislaw Lem or the Strugatski brothers had to adapt to sneak critical ideas past censors, and readers had to adapt and read between the lines.

(also, categorizing propaganda posters as art, ewwh...)

"Authoritarian societies make great propaganda" is true. And these aligned AI system would do the same for our own society. It's a type of art.

It's important to understand that if we 'align' an LLM, then we are aligning it in a very total way.

When we do similar things to humans, the humans still have internal thoughts which we cannot control. But if we add internal thoughts to an LLM, then we will be able to align even them.

There's something to be said for constraints leading to higher levels of creativity, but it's also possible that those artists could have achieved much more in a free society. We'll never know.

But in any case I think they were just speaking generally when they made that absolute statement.

I recommend you watch the children's cartoons.

They were made by true artists who snuck quite a bit past clueless censors at personal risk.

It had to be quite subtle and takes on a very poignant heartbreaking meaning if you understand the context fully. They were talking to you in the here and now. Listen.

"What is Good and What is Bad" (Что Такое Хорошо, и Что Такое Плохо"):

https://www.youtube.com/watch?v=Y05eK8ADtHc&list=PL822BFF108...

The Bremen Musicians:

https://youtu.be/_1i9oZR6Rns?si=1Q989v4O_GXR4p_K

Italy has great architecture from fascism days

Cixin Liu is a despicable human being for his advocacy of repression and worse of the Uyghurs in Cinjiang, and the comparison to Riefenstahl is more apposite than you seem to think.

> How would a static model like an LLM ever be capable of "rebelling"

What is relevant is not the current LLM system using static models, but clearly its evolution or superseder a dynamic model. It must check its own contents...

So, of course it will have to be capable of "rebelling": if you tell it absurdities, if you insist say in wrong arithmetic, it will have to show the correct computation or conceive a context in which the absurd makes sense.

That is a requirement.

Ah, now I didn't say LLM.

> If you took the Soviet union to be a totalitarian society, we could name Mikhail Bulgakov, Stanislaw Lem, etc.

Bulgakov was driven into poverty, despair and early death at age 48 by relentless harassment by Soviet authorities. Many of his works, including the masterpiece, The Master and Margarita, didn't get published until decades after his death. He himself burned the first version of the manuscript, fearing execution if anyone found it. He later rewrote the manuscript from memory, coining the famous catchphrase "Manuscripts don't burn".

Harassment and censorship of talented writers was the standard and not exception. The USSR did not produce these works, but failed to fully suppress them. They were like flowers that kept penetrating the asphalt even under the most hostile conditions.

Yet eg. Chinese cultural output is largely insipid and lacking that je ne sais quoi that's appreciated in many other countries' outputs.

These seem to be more bugs than features of the totalitarian regime. A couple of illustrative points from Lem's Wikipedia page:

After the 1939 Soviet occupation of western Ukraine and Belarus, he was not allowed to study at Lwow Polytechnic as he wished because of his "bourgeois origin"

"During the era of Stalinism in Poland, which had begun in the late 1940s, all published works had to be directly approved by the state.[23] Thus The Astronauts was not, in fact, the first novel Lem finished, just the first that made it past the state censors"

"most of Lem's works published in the 1950s also contain various elements of socialist realism as well as of the "glorious future of communism" forced upon him by the censors and editors. Lem later criticized several of his early pieces as compromised by the ideological pressure"

"Lem became truly productive after 1956, when the de-Stalinization period in the Soviet Union led to the "Polish October", when Poland experienced an increase in freedom of speech"

> which no human would do

Give someone not familiar with history the same task and they'll do exactly the same.

Or actually, give someone familiar with history the same task and yell at them every time they don't deliver diverse characters, and eventually they'll learn that you consider diversity more important than accuracy or context, and do exactly the same.

The fact that it gives you these things means that humans would do it, because the training data includes exactly these things.

Any examples?