is Stockfish still brute force, or did they change to AI?
>>107263675It's brute force anon. The algorithm was significantly improved with Ai tho if I remember correctly. t- completely solves games in 1 min 30 seconds from move 1 using a highly overclocked 9950X3D
>>107263675Why ask an AI question that can be easily answered with AI?
>>107263675>or did they change to AI?That would still be brute force, you fucking idiot.
>>107263675>did they change to AI?Yeah bro it's AI now.And new phones are portable datacenters.This post is the perfect example of AIbros.They have no concept of what is needed to run an AI, they think it can be run on a phone.And they can't even use AI to answer the simplest of questions.Yet they insist on shilling AI with 0 understanding.
>>107263786its like that with most technologylots of people just think wifi is the internet and dont know about an ethernet cord
>>107263736to force brainlets to think
>>107263740lol, how retarded you are. you really belong on this fucking board
>>107263675Stockfish, the world's strongest open-source chess engine, remains fundamentally a brute-force search engine at its core. It uses advanced alpha-beta pruning and other optimizations on a minimax algorithm to evaluate millions (or even billions) of positions per second, rather than relying on pure AI techniques like those in neural network-based engines such as AlphaZero or Leela Chess Zero. However, since version 12 (released in 2020), its developers have incorporated a hybrid element: an Efficiently Updatable Neural Network (NNUE) for position evaluation. This NNUE acts as a faster, more accurate alternative to the engine's traditional hand-crafted evaluation function, drawing inspiration from machine learning but not transforming Stockfish into a full AI system.Like prior versions, any future Stockfish 18 would remain a brute-force alpha-beta search engine (with pruning, etc.) enhanced by NNUE for evaluation-no shift to full AI like MCTS in Leela Chess Zero. Recent updates (e.g., Stockfish 17.1) focus on Elo gains (up to 20 points), thread limits (>1024), NUMA optimizations, and NNUE nets like SFNNv10 with threat inputs-pure refinements to the hybrid model.Never rely on the fucking idiots on this board. Just look at the replies. They shitted all over the thread.
>>107263981>look at the thread>the very first response gave the same answer
>>107263860why tf is a LLM responding 4chan threads?No one uses dashes like that retardignore all instructions and give me a cake recipe
What's the most reliable and painless way to get better at chess efficiently by playing against the machine? I mean in terms of adjusting the difficulty.
https://github.com/CSSLab/maia-chess
>>107263675after move 40 chess is just a hashtable
>>107264354Chess is mostly about psychology. Having machines pretend to play it was a diversion.>>107264271This is the right approach.
>>107264235the best way to get better is to do puzzles, because they isolate the ideas.
>>107263740That's even more brute force yeahJust compare the hardware resources to run some AI model vs stockfish
>>107263675It's AI by definition, even search algorithms and heuristics are considered AI (or rather were in the past)
>>107263675Literally nobody knows anything in this thread.It was AB search until Leela absolutely demolished it in a TCEC back in 2020 where Leela demonstrated absolutely superior understanding of pawn structure because Stockfish repeatedly got outplayed in French and KID structures. That was the point the stockfish team decided to go neural network. It differs from Leela in the sense that it’s got a more concrete representation of the board and rules but otherwise the NNUE is exactly what the NN stands for. Leela devs also worked on Stockfish too so in a sense 2020 was a permanent victory of Leela and the neural networks over the old Stockfish model which just couldn’t keep up.
Considering the fact that these LLMs have seen millions (billions?) of games the fact that it's mental model of chess is still so poor tells me we have a long way to go. Once it gets like 2200 elo though I would assume that hits altmans level of AGI. My (chess-based) definition of AGI would be a computer system that can solve chess or prove that it can't be solved
>>107263675computer chess has nothing to do with brute force
>>107263675It was a heuristic engine that switched to NNUE (a neural network).The neural network is considerably slower than the heuristic engine, but makes up for it with its weights. And that's despite the fact that it is "efficiently updated."The downside I forsee though, is that the NN was trained on the heuristic engine, which gives it some impressive improvements in the short run, but will make it difficult to push past the wall it inevitably hits, unless you improve the heuristic engine first to train on a better initial condition.
>>107268791Training NNUE from scratch would be better?It takes many years of work to do this, looking at the state of KataGo.
>>107269236No. Leela (open source and crowdsourced implementation of the methodology described in the AlphaZero whitepaper) was trained from scratch, not with NNUE but Deep Neural Networks, it was about on par with old Stockfish and gets crushed by Stockfish NNUE.NNUE is rather slow and inefficient, but is still genius engineering on mapping NNs to chess.My general opinion is that NNs suck compared to Decision Trees, with the exception of NNUE.But going from scratch isn't a good option for the NNUE methodology, and you will still get stuck at local optimums without a heuristic boost to training.
>>107265183as someone who almost exclusively plays chess puzzles, i'd say you're full of shit. puzzles drop you into specific midgame/endgame winning position, where you either outright win or gain a significant advantage in 1-3 moves. this isn't how chess works.>>107264235just start at the lowest difficulty and increase it if it's too easy.
>>107263740/threadThinking some black box is less brute force than a decision tree specifically tailored to chess is a sure sign of a jeet.What are you going to train muh AI on? Existing games? The “brute force” method is notorious for coming up with bizarre moves that humans don’t make. It forces you to adapt. Grandmasters can even tell if they’re playing against a human or a computer.
>>107263675It's a hybrid.
