>AI can officially play chess better than the world champion
It's so over. AI is going to take over all white-collar jobs by 2004
 It's All Fucked Shirt $22.14 |
 Thalidomide Vintage Ad Shirt $22.14 |
It's All Fucked Shirt $22.14 |
>AI can officially play chess better than the world champion
Underrated post
That wasn’t actually AI though. Traditional chess engines are programmed with pre-defined heuristics, whereas an engine like Leela or AlphaZero learned everything through pure self-training.
And yes it is a great analogy because when AI has a clear way of analyzing its quality (by looking at win rates, in this case), it has the ability to absolutely embarrass the greatest chess grandmasters. With something like programming, it’s not so easy for the AI to just train itself. It needs human feedback, which takes a lot longer to train it. It will still eventually surpass us, especially if we develop a way for it to automate the training process without our needing to give feedback.
Its just pattern recognition. In a pre-defined game like chess, its easy to react to moves on the opponent.
I think the computer will win at chess, but its not surpassing us in intelligence as I think you insinuate.
Literally everything is just pattern recognition. They were all saying EXACTLY what you’re saying about chess engines, that they will never beat humans, and even when Kasparov lost he thought the programmers had cheated somehow. Humans don’t want to think that their brains are not much different from computers, so we make up excuses like “it doesn’t actually create anything” or “it just copies what it’s learned in the past” WHICH IS EXACTLY WHAT HUMANS DO. We just had millions of years of evolution to train our brains, whereas AI has only been training for a few years at most. The chess engine AlphaZero learned everything it needed after only 4 hours of playing against itself
That's an interesting point to say that the "human model" was created through evolution, and probably correct.
yes, fundamentally it’s the same. AI is just much faster because it can mutate at a faster rate and isn’t limited by the need to reproduce. So AI will continue to evolve and there will be a sort of “species” of AI from which only the best models will survive, and from there those will be modified and the race will continue. The ultimate goal of course is to have a model that can train itself on any topic in the same manner that the chess engines can. And maybe it would be able to modify its own code and create better versions of itself.
get your GED son
I’m telling you exactly how the future will unfold.
You're right, AI is amazing at Chess. Us mere mortals can't comprehend their plays
Dont listen to this guy, 2000 is gonna be THE year for tech. Make sure to buy up as much tech stock as possible!
Taking this opportunity to shill this movie.
It was made 10 years ago but the commentary on AI is still oddly relevant today. There's people discussing whether or not AI can surpass humans, there's some guy from the Pentagon infiltrating to assess the military applications of AI, and there's even a scene with a chatbot.
I know it's a work of fiction, but the deeper I research history the more I'm really surprised by how timeless these discussions on AI are. Not really sure what to take away from it, since obviously we are making some incredible advancements today.
That AI is fundamentally different from modern AI.
Back then, you had the actually program the AIs. You had to know what you wanted them to know and actually specify it. It couldn't do stuff you didn't specify.
Now we just hand it a loss function and it spits out an incoherent matrix of floating point numbers that just works. We have no idea what's going on inside so it surprises us when a text predictor can do everything ChatGPT can do.
>Boomers think this won't happen
>It's just a game!
cool and driving a car is just a game too. Green light go brrr red light stop
self driving cars by 2001
>ARRRRGH SAVE ME CHESS AI HAS COME TO KILL ME
>muh data center supercomputer that consumes 10 gigawatts and costs $5 billion can beat a human (whose body and brain were given to him for free by God and only consumes 100 watts)
Meanwhile it still can't draw a hand
What is going on ITT? Chess computers have been bearing people for decades. (Deep Blue vs. Kasparov) Stockfish works using lookahead (with some pruning for speed) and then evaluating the position according to heuristics at every branch, picking the branch that leaves the opponent with the worst best option (min-max).
The evaluation function is based off of trial and error and understandings made by humans (rooks are worth more late game, pawns gain value for each rank they advance, a blocked bishop is worth less than an open bishop).
Leela and AlphaZero use real ML and are purely/mostly looking at patterns rather than rules. I'm not as familiar with their implementation. However they were very purpose built and trained for a very long time.
Current text synth AI is cool and good at text synth/transformation but it sucks at chess. Not only does it lose, it has no notion of correctness and will soon make illegal moves once a game is out of the book moves.
This text synth AI will never be able to understand something. Humans will always be able to outplay it since it has no ability to analyze a position or even understand the state of the board. It is simply predicting the next word in a sentence.
One tip, Stockfish running on a Raspberry Pi will crush a GM today. No supercomputer needed.
I thought chess was a solved game like checkers? AI should never lose given the size of modern models.
Nope. Chess is an extremely complex game.
No they keep creating chess bots that are smarter than the last version despite them being beyond human long ago
AI means nothing unless it can physically locate and move the chess pieces on its own for any chess board in any setting just like a human could