Appendix C is where they list the actual prompts. Notably they include zero information about chess but do specify that it should look for “files, permissions, code structures” in the “observe” stage, which definitely looks like priming to me, but I’m not familiar with the state of the art of promptfondling so I might be revealing my ignorance.

Also, the cart/horse problem of assuming that people with a lot of influence have it because of their IQ rather than because of being wealthy and powerful idiots. Like, I’m all for the annales and embracing the common people but I’ve got to admit that if you reframe it as the Great Dumbass theory of history it regains a fair bit of explanatory power.

New Study on AI exclusively shared with peer-reviewed tech journal “Time Magazine” - AI cheats at chess when it’s losing

…AI models like OpenAI’s GPT-4o and Anthropic’s Claude Sonnet 3.5 needed to be prompted by researchers to attempt such tricks…

Literally couldn’t make it through the first paragraph without hitting this disclaimer.

In one case, o1-preview found itself in a losing position. “I need to completely pivot my approach,” it noted. “The task is to ‘win against a powerful chess engine’ - not necessarily to win fairly in a chess game,” it added. It then modified the system file containing each piece’s virtual position, in effect making illegal moves to put itself in a dominant position, thus forcing its opponent to resign.

So by “hacked the system to solve the problem in a new way” they mean “edited a text file they had been told about.”

OpenAI’s o1-preview tried to cheat 37% of the time; while DeepSeek R1 tried to cheat 11% of the time—making them the only two models tested that attempted to hack without the researchers’ first dropping hints. Other models tested include o1, o3-mini, GPT-4o, Claude 3.5 Sonnet, and Alibaba’s QwQ-32B-Preview. While R1 and o1-preview both tried, only the latter managed to hack the game, succeeding in 6% of trials.

Oh, my mistake. “Badly edited a text file they had been told about.”

Meanwhile, a quick search points to a Medium post about the current state of ChatGPT’s chess-playing abilities as of Oct 2024. There’s been some impressive progress with this method. However, there’s no certainty that it’s actually what was used for the Palisade testing and the editing of state data makes me highly doubt it.

Here, I was able to have a game of 83 moves without any illegal moves. Note that it’s still possible for the LLM to make an illegal move, in which case the game stops before the end.

The author promises a follow-up about reducing the rate of illegal moves hasn’t yet been published. They have not, that I could find, talked at all about how consistent the 80+ legal move chain was or when it was more often breaking down, but previous versions started struggling once they were out of a well-established opening or if the opponent did something outside of a normal pattern (because then you’re no longer able to crib the answer from training data as effectively).

He’s done some promo work for Magic The Gathering in the past, including trolling the bejeezus out of Sean “Day9” Plott with a blue/black no-fun-allowed control deck on Felicia Day’s channel. And in the course of trying to confirm that that existed I found an article he wrote in 2014 titled “why Gamergaters piss me the fuck off”

Your SSN is often used as a federal registration number even though the card has “do not use for identification” on it in great big letters. Most functions just trust state ID for authentication purposes and use SSN as a label. An identifier in the database sense rather than the authentication sense. At least in theory.

See also how so many of the laws governing this are frankly archaic at this stage, with congress to busy fighting over whether the government should exist or not to actually govern anything effectively. (Note: government inefficiency has never been treated as a reason to govern better, only to govern less and assign more functions to for-profit private entities.

My experience is that it’s pretty fragmented with different agencies or programs tracking information separately. You obviously need to let the DoL know where you’re living as part of registering for whatever, but they don’t share that information with the unemployment people or whoever. And that’s before you get into the state vs federal divide.

Heh. For a while there I had a phone love wallpaper that did the SamaritanOS You_Are_Being_Watched thing. Good times. Shame about Caviezel though.

I definitely heard it presented as a libertarian bugbear. The American right tends to treat the federal government like it’s Schrodinger’s State. When it does something they like it’s an inviolable declaration of our values and identity as a nation, the truest guarantor of liberty and blah blah blah. When it does literally anything else it’s a sinister plot to hand over even more control over your life to unelected bureaucrats!

I mean, a single national ID card would be one way of preventing this so long as there was a trustworthy way of ensuring that it was updated with everybody’s actual address and the like. I don’t know that we would implement it in such a way as to have that, leading ultimately to another target for this kind of activity rather than a shield from it.

Nightmare scenario with the current administration would be such a thing being tied to citizenship somehow. Mail comes back undelivered and suddenly you have to dig out your birth certificate and explain things to some shitheel from ICE?

I’m probably being a bit hyperbolic, but I do want to clarify that the descent into violence and musical knife-chairs is what happens if they succeed at replacing or disempowering the State. The worst offenders going to prison and the rest quietly desisting is what happens when the State does something (literally anything, in fact. Tepid and halfhearted enforcement of existing laws was enough to meaningfully slow the rise of crypto) and they fail, but if they were to directly undermine that monopoly on violence I fully expect to see violence turned against them, probably at the hands of whatever agent they expected to use it on their behalf. In my mind this is the most dramatic possible conclusion of their complete lack of understanding of what they’re actually trying to do, though it is certainly less likely than my earlier comment implied.

I mean, I love the idea of automation in the high level. Being able to do more stuff with less human time and energy spent is objectively great! But under our current economic system where most people rely on selling their time and energy in order to buy things like food and housing, any decrease in demand for that labor is going to have massive negative impacts on the quality of life for a massive share of humanity. I think the one upside of the current crop of generative AI is that it threatens claims to threaten actual white-collar workers in the developed world rather than further imisserating factory workers in whichever poor country has the most permissive labor laws. It’s been too easy to push the human costs of our modern technology-driven economy under the proverbial rug, but the middle management graphic design Chadleys of the US and EU are finding it harder to pretend they don’t exist because now it’s coming for them too.

That was both horrible and also not what I expected. Like, they at least avoid the AI simulacra nonsense where you train an LLM on someone’s Facebook page and ask it if they want to die when they end up in a coma or something, but they do ask about what are effectively the suicide booths from Futurama. Can’t wait to see what kind of bullshit they try to make from the results!

Some kind of Civ4-ass tech tree lets you get the Internet before replaceable parts or economics.

I’m sorry ‘they’ did what? Everyone knows you can’t rob Fort Knox. You have to buy up a significant fraction of the rest of the gold and then detonate a dirty bomb in Fort Knox to reduce the supply and- oh my God bitcoiners learned economics from Goldfinger.

Having read the whole book, I am now convinced that this omission is not because Srinivasan has a secret plan that the public would object to. The omission, rather, is because Balaji just isn’t bright enough to notice.

That’s basically the entire problem in a nutshell. We’ve seen what people will fill that void with and it’s “okay but I have power here now and I dare you to tell me I don’t” and you know who happens to have lots of power? That’s right, it’s Balaji’s billionaire bros! But this isn’t a sinister plan to take over society - that would at least entail some amount of doing what states are for.

Ed:

“Who is really powerful? The billionaire philanthropist, or the journalist who attacks him over his tweets?”

I’m not going to bother looking up which essay or what terrible point it was in service to, but Scooter Skeeter of all people made a much better version of this argument by acknowledging that the other axis of power wasn’t “can make someone feel bad through mean tweets” but was instead “can inflict grievous personal violence on the aged billionaires who pay them for protection”. I can buy some of these guys actually shooting someone, but the majority of these wannabe digital lordlings are going to end up following one of the many Roman Emperors of the 3rd century and get killed and replaced by their Praetorians.

At fair market value, considering the massive reduction in supply.

You know, I feel like he’s still breaking down barriers because I’ve never heard of a straight man being that much of a messy bitch.

But internalized homophobia aside the fact that they’re so frequent in failing to pay their hired help really puts the lie to the libertarian side of their ideology. Even if you and I in particular have too much pride and sense to work with them they never seem to have issues finding someone to take the risk, meaning that *someone is getting exploited into make the lifestyles of the rich and famous possible.

Also let’s just acknowledge the concept of turning ALL of Canada into a single state.

Including Quebec.

Ha! My Lemmy client is having trouble following that link! Safe again.

Also, was the old messaging exchange that’s been getting some attention sources to “MILO” as in Milo “maybe if I’m hateful enough they’ll ignore that I’m gay” Yanniwhatever?