People that still wonder if the generative AI models can actually "think" should try playing chess against them. ChatGPT just beat me by conjuring a third rook and then moving it diagonally past 2 pieces for checkmate. No human could have the creativity and intelligence to pull that off - ThreadSky

steinkobbe.bsky.social • 176 days ago

People that still wonder if the generative AI models can actually "think" should try playing chess against them. ChatGPT just beat me by conjuring a third rook and then moving it diagonally past 2 pieces for checkmate. No human could have the creativity and intelligence to pull that off

Comments

anywhy.bsky.social•176 days ago

I would simply use the Deathnote to give my opponent's king a heart attack.

dansedelectre.bsky.social•176 days ago

I recall reading something once about an AI model running a simulated race that won instantly by making the runner taller than the track and falling over

sparksbet.bsky.social•176 days ago

Weirdly, the models that do things like that (typically reinforcement learning models) have a better understanding of the rules they're finding loopholes in than ChatGPT does of the rules of chess.

orman.bsky.social•176 days ago

Sorta? They do at least work rigorously within those rules, ChatGPT is just spitting out "things that look like algebraic notation" without any check against the actual rules of chess

sparksbet.bsky.social•176 days ago

Yeah that's more or less what I mean, in reinforcement learning, the model gets rewarded and punished for doing certain things, which actually "learns" the rules, whereas ChatGPT is literally just a language model with no chess model.

plotjes.bsky.social•176 days ago

Groaned so hard at the ACX post that people can not tell AI art ftom human-made art because all the prompts were 'imitate something that Van Gogh painted' and whoa! It looks a lot like something Van Gogh could have painted!

Meanwhile it still fumbles any serious research.

plotjes.bsky.social•176 days ago

Ask it a question like 'give me some opinion pieces from the NYT/WaPo/WSJ/USAToday that argued against the Iraq invasion before the invasion started' and you get like a completely unrelated Slate article from 2009 on how Hillary Clinton is the future of the Democratic party.

joqatana.bsky.social•176 days ago

Don’t ask it anything.

steinkobbe.bsky.social•176 days ago

As long as we're fine with nothing truly new ever being created again then we're ready for the machines to take over art

plotjes.bsky.social•176 days ago

When AI took off two years ago there were so many 'what if Ghostbusters was directed by Wes Anderson?' 'what if Caddyshack was directed by Wes Anderson?' 'What if Hellraiser' etc

And those are fun things to do if you're in the first year of film school but eventually you gotta make your own shit!

mwebber.bsky.social•176 days ago

You got Kobayashi Maru'd by a dumb machine!

drukerverle.bsky.social•173 days ago

Have you seen the moves that horse thing can do? Runs off with queen and then deletes the board.

ryankeelytm.bsky.social•176 days ago

Reminds me of Kelly Flannigan winning the obstacle course tricycle race group date by skipping the obstacles. Loved Pilot Pete’s season of the Bachelor….

boredompeejeterson.bsky.social•176 days ago

You should play against little kids who don't know the rules more

darksidestraxus.bsky.social•176 days ago

Reminds me of AI bros who brags about how nobody can tell the difference between AI art and regular art when AI art has that waxy quality to it.

radianthostility.bsky.social•176 days ago

Yeah. They're just telling on themselves. They're not meaningfully literate so they don't know how ridiculous their claims sound to people who are.

tractusb.bsky.social•176 days ago

It frustrates me to no end how a real artist's style was stolen and mass generated by AI that the original art gets mistaken for that filth.

Gen-AI may finally imitate real art one day, and the tragedy will be reducing a potentially helpful tool to stealing and denying credit from hard workers.

ughnaught.bsky.social•176 days ago

There are so many AI videos on YT right now of Christmas scenes with Christmas music played over them. They all have a horrible uncanny valley feel with humans moving inhumanly, bizarre design choices, and a Rockwell-esque haze over it all.

renaigh.bsky.social•176 days ago

Sleight-of-hand check crit success.

jvdrieth.bsky.social•176 days ago

I won't share with you what happened when I played this weird Counterstrike x Chess hybrid until they've rebuild my village and returned my family

onelifestand87.bsky.social•176 days ago

@mbellcoop.bsky.social for your next game

mbellcoop.bsky.social•176 days ago

Aye I've seen some of it's games before, they really hurt your brain watching them

prepostroff.bsky.social•176 days ago

this is such a bizarre metric to have decided to use for this? Computers have been annihilating us at chess for many years now and I don't think anyone would say that means those programs are "thinking" either?

steinkobbe.bsky.social•176 days ago

Yeah they're not thinking they're designed for a specific task and most people understand that about chess engines

prepostroff.bsky.social•176 days ago

They could easily rig an LLM to have access to one of those chess algos and just implement it whenever it encountered a chess match. Would that LLM then be “thinking”? It’s useless as a metric of what would or would not define “thought” from a computer, you just said a thing that *feels* true.

steinkobbe.bsky.social•176 days ago

👍