ChatGPT's commentary on its moves is great. "I captured your pawn on d5 with my knight, creating a double attack on both your queen and the newly captured pawn." Even top human grandmasters rarely see the potential of attacking pieces that have already been captured.

'as a GM unlike Levy' beautiful

I like how Ben is calculating possible comments and refuting them before they even happen. He's like an engine.

"And then ChatGPT played Bb5+ ☝️AGAIN☝️"

Leela is free online! But there’s no reason to get it over Stockfish, and, like you said, any of the top engines are more than strong enough to help the gawking rabble. Ben is right yet again.

this was a really helpful lecture actually. I watch a lot of your lectures on openings and famous players and whatnot, but never actually learned how to study chess or tips on how to do that... so thanks!

I am a robotics engineer and use ML in my job. Saying that chatGPT is an AI for everything is incorrect. It's a language model, it is basically a sophisticated parrot (although a parrot is probably smarter in terms of general intelligence). It can't play chess because it doesn't understand what chess is. There's no actual intelligence behind it, it's just a text predictor that is good and figuring out what words go well as answers to the words you're feeding in.

ChatGPT is not "an AI for everything". It's a language model. It's good at producing text in a way that is similar to things people have written on the internet. That's it.

Awesome you landed a sposor. Hopefully you land many more and keep this one. This was a great lecture Ben don't sell yourself short. I learned more on this lecture than I did for a while.

Obviously Leela is free and everybody knows that why don't you know it?

In chessbase, when you were looking to create a db with those Xiong games, does selecting one, then hitting Ctrl+A select them all? Hoping to teach you something for once Ben!

Great introduction to these tools, thanks Ben. I’m 2200 and nearly all of this was new to me.

Thanks Ben, for the Chessable info. Great lesson, overall. Looking forward to more.

Thank You, Ben; .. absolutely Fascinating video 😊

thank you for the lecture gm ben finegold

Somehow I would be interested in a high resolution picture of Mr. Finegold's shelf to see what chess books he used. Could only recognize a few. :)

Simply Brilliant.

When will I be able to purchase Bark like a Grandmaster?

Babe wake up new Fingold video just dropped

I think the important thing about analyzing with an engine is the fact it will point out lines you hadn't considered, especially much better lines and for both sides. And very quickly at that. The engine doesn't need to be all that much better than you for this purpose.

In the explorer it doesnt show the most common, but rather the position. If there are transpositions its hard to tell what move is actually common. For example try Nf3 Nc6 Ng1. It will say Nb8 is the most common move, which is obviously nonesense

Congrats on your sponsor Ben

Surprisingly, almost nothing said was wrong. Stockfish is indeed much better than Leela, which is much better than AlphaZero was back when it still existed (it was short-lived like Deep Blue). Also Stockfish and Leela are indeed both free. The one thing he missed is that modern Stockfish ("Stockfish NNUE") now incorporates neural nets as well, although to a lesser degree than the other two.

Is there a bongcloud course on chessable??

Go Ben!

4:38 obviously it played the OTHER bishop to b5 smh

StockFish is likely the best based on the most objective measures available namely 100 or ideally more games in various forced openings and free openings. At very small time settings, there might be special engine builds that could do better e.g. milliseconds or microseconds. That it is free and open source and has advanced heuristics specially for evaluating chess postitions also neural networks and a large developer community is even better.

If you give gpt the pgn, it's less likely to make illegal moves. It typically plays like stockfish 14

Maybe you should ask it to play legal moves

Hey finegold I got a ?. How do you blunder on both sides of the board 😊

I like Lichessdotorg because it is free. Thanks for the video Ben.

Great Engines of the Past

If I understood it correctly, chatgpt has an issue with something called object permanence. It dosen't understand, that there were physical objects simulated who have implications on the movement of the pieces. Chatgpts core is a statistical word prediction algorithm. By this, it generates a reaction based on the last interaction, partly independently of what had happened before. It also explains partly the moves it played against Ben. The most played opening is e4 going into the italian or spanish. So statistically the e pawn has already moved when you play Bb5. In the same manner, usually - most likely - there is a knight on c6, so there is no issue with the bishop being attacked by a pawn. The algorithm dosen't comprehend, that the situation has changed and how this influences the required response. It's a statistical tool, not a linear.

go ben!

i'm always fascinated by the fact that not only "best move" is not the best move, but that even a move better than "best move" may not be the actual best move

do you remember our conversation about chess AI forty years ago? hoy hoy digging up bones send me the bill

* many comments * and so forth

The engine comparison is like asking Magnus or Wesley So to analyse my game. Just because Wesley So's rating is lower, doesn't mean his analysis has less value to me.

Leela is free and quite strong but you need a good GPU (the CPU-only version sucks)

I love you Ben

Me: Hey. Here is a chess game a played. What do you think? ChatGPT: Maybe chess is not your game. Try checkers. Stockfish: What he said.

Hmmm... it seems Ben needs more comments. Here is one more.

chatgpt was fun when it first came out, but if you spend any time with it you find out it can;'t do anything. It is just writing random words that sound somewhat coherent.

It already played d4, so I don't have to do anything 😂😂😂

Im not an expert, but ChatGPT is just working with a practically endless source of text and doing analysis on that huge amount of info, and coughs up stuff that *look like* a reasonable answer to your questions. ChatGPT playing chess is chaotic because it doesnt actually play chess, it just coughs up random algebraic notations, because for its text-analysing algorithms, chess is nothing more than random letters and random squares named after each other.

Just a little history about Alpha Zero. The type of AI that it uses can have its calculation time accelerated on graphics cards. Long story short, a graphics card has hundreds of tiny processors (whereas your processor has maybe 2-20), and it helps that kind of AI crunch numbers to find an answer. The trick the Open AI team did when they beat Stockfish, which only runs on CPUs, is they had their famous matches with Stockfish using a CPU (way less computing power per unit time) versus Alpha Zero running on a GPU (way more computation per unit time). In that sense, Stockfish had a clear compute disadvantage. This would be like having Stockfish with 30 minutes of compute time play Stockfish with 10 minutes of compute time and then be like, "Wow, Stockfish is better than Stockfish." I have to be fair and admit Alpha Zero uses a different technique than Stockfish, and it was still impressive. I'm just not sure if, with the computation the same, it would have actually beaten Stockfish. Alpha Zero got a lot of wins in closed positions where having more compute lets the engine see further. More or less, they structured the tournament to where Alpha Zero would surely win. There are chess engine tournament organizations, and they run Leela Zero on CPU mode (even though it can run on a GPU to get accelerated just like Alpha Zero). It still does a great job as it's a strong engine, but Stockfish is still king over it. Plus, Stockfish integrated a bit of that AI magic in newer versions, making it even stronger. That's what that NNUE thing is all about. That's stockfish using an AI model to evaluate positions. It's a much more lightweight AI model than the one Alpha or Leela Zero use, but it still outperformed the old evaluation model that was handwritten by humans.

yes, leela chess zero, lc0, is free and open source

Why is Bens chair vibrating so much? What is happening?

Would you have been able to tell Chat GPT that only knights can jump over other pieces so it would learn and not try to use the bishop incorrectly?

chat gpt is a language module, still you can just tell him, hi look you can not make this move because xxx and it will try and correct the move

I feel like Ben is trying to come up with some new Finegold quotables? "You're wrong, but even if you were right, you're still wrong... and "You're so bad at chess, you're opinion matters even less than when I didnt care"...

Ben could teach chess to chapgpt. Then you'd really show him!

GM Finegold is completely wrong. He is the leading authority on the things he's talking about.

It's not good at playing chess with you. But sometimes I ask it to explain in depth why a move was a blunder, which is helpful. The engine is not good at explaining "why" and I'm not good at memorising moves if I don't know the why.

Stockfish stronger faster harder better

Chat got cheats moves? Should be a great game against👋👋 Nieman.

Continuity is a problem with all of these AI's.

Ypu cannot mass.refund on chessable. Some ppl tried to be clever. I could have had 3 courses refund at once but I wouldn't rry much more. But you need months to go threx some of those courses anyway. So pick a few you like and dont be afraid refund part of them.

This comments section isn't the worst at all! Ben you are a liar and a fraud for telling us it would be bad!

Why is your chair shaking?

omg watching this drunk is just great also you're wrong about a lot of things

Ben? Why is your belly dancing?

This is the worst comment on any video you ever made!

Why Chat GPT is so bad at chess? Because he is talking too much!

Youre wrong about a lot of things

your sponsor seems r**arded tbh

"very funny at the beginning" ah, i see you will be using chat gpt The real issue here is that this alleged AI is doing absurd things ALL THE TIME. Here, we can see this, because we know how to play chess. But many, many people will be ASSUMING that this alleged AI is infallible, and will be using it to produce absurd results in fields that they don't know anything about, and will be treating those results as gospel.

Ben is too late to the party