TechTakes

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).

this was so shocking that at first I thought it must be satire https://youtu.be/VwlBwyJVEfw

New piece from Brian Merchant: 'AI is in its empire era'

Recently finished it, here's a personal sidenote:

This AI bubble's done a pretty good job of destroying the "apolitical" image that tech's done so much to build up (Silicon Valley jumping into bed with Trump definitely helped, too) - as a matter of fact, it's provided plenty of material to build an image of tech as a Nazi bar writ large (once again, SV's relationship with Trump did wonders here).

By the time this decade ends, I anticipate tech's public image will be firmly in the toilet, viewed as an unmitigated blight on all our daily lives at best and as an unofficial arm of the Fourth Reich at worst.

As for AI itself, I expect it's image will go into the shitter as well - assuming the bubble burst doesn't destroy AI as a concept like I anticipate, it'll probably be viewed as a tech with no ethical use, as a tech built first and foremost to enable/perpetrate atrocities to its wielder's content.

Occasional sneerclub character Nate Silver is bluechecking again.

haha, it's starting to happen: even fucking fortune is running a piece that throwing big piles of money on ever-larger training has done exactly fuckall to make this nonsense go anywhere

The New York Times Pitchbot enters our territory:

We wanted to understand the future of AI. So we talked to three Hawk Tuah cryptocurrency investors at a White Castle in Toms River.

Not really a sneer, nor that related to techbro stuff directly, but I noticed that the profile of Chris Kluwe (who got himself arrested protesting against MAGA) has both warcraft in his profile name and prob paints miniatures looking at his avatar. Another stab in the nerd vs jock theory.

google's on their shit again

can't sneer it properly just yet, there's a lot

Media companies continue to impress me with their hard-hitting investigative boots on the ground reporting

Write a brief article titled "ICE Prosecutor Linked to Anonymous White Supremacist X Profile: Report"

The most naked attempt yet at allowing billionaires to live on without the rest of us.

What infuriates me the most, for some reason, is how nobody seems to care that the robots leave the fridge door open for so long. I guess it's some form of solace that, even with the resources and tech to live on without us the billionaires still don't understand ecosystems or ecology. Waste energy training a machine to do the same thing a human can do but slower and more wastefully, just so you can order the machine around without worrying about it's feelings... I call this some form of solace as it means, even if they do away with us plebs, climate change will get'em as well - and whatever remaining life on Earth will be able to take a breather for the first time in centuries.

Deep thinker asks why?

Thus spoketh the Yud: "The weird part is that DOGE is happening 0.5-2 years before the point where you actually could get an AGI cluster to go in and judge every molecule of government. Out of all the American generations, why is this happening now, that bare bit too early?"

Yud, you sweet naive smol uwu baby~~esian~~ boi, how gullible do you have to be to believe that a) tminus 6 months to AGI kek (do people track these dog shit predictions?) b) the purpose of DOGE is just accountability and definitely not the weaponized manifestation of techno oligarchy ripping apart our society for the copper wiring in the walls?