Lichess

Um, s- I bounce around.

So I, I recently started playing on Lichess.

Okay.

Um, because, um, I listen to Neil Bell on, uh, Chess Angle.

I- one of the things that I find most interesting about the two servers is that on chess.com, you cannot get a 5|3 game.

Or if you do, you have to wait a long time.

On Lichess, it's one of their defaults, and you click it, and bam, you immediately have a game.

So it's kind of interesting that on the different servers, different time controls are preferred by the players.

Why anyone would bother? Honestly, for ordinary online chess, they wouldn't need to.

Lichess and Chess.com already do a fine job storing move histories and settling disputes because they're trusted, centralized referees.

ZK chess earns its complexity in rarer situations, trustless stakes.

If two strangers want to wager crypto on a game with no platform holding the funds, something has to adjudicate the outcome on-chain without a human in the loop.

Chess ratings are public and computed by transparent, well-documented algorithms, Elo, Glicko.

The strongest engine, Stockfish, is free and open source.

The most popular serious platform, Lichess, is open source and nonprofit.

A lot of what Web3 promises other industries, open access, verifiable results, no rent-seeking gatekeeper, chess arguably got there first without a blockchain.

I think now it's maybe not so easy to do that, but I think for someone today you have to...

I think you have to play as much as you can, and I don't really mean playing, like, blitz games on Lichess or Chess.com. 'Cause when you play...

I think when you play online, it's too easy to just not really take it seriously.

Yeah.

And David, are you, like, are you following pro chess? I mean, obviously, you keep up with the opening theory and stuff, but are you excited for the world championship, and do you, uh, play through games every day and stuff like that still?

Uh, yeah, I mean, I, I tend to look, um, at what's going on.

Um, like, I'll go on Lichess, uh, broadcast and see what tournament's going on.

I often will follow American tournaments-

Since I began teaching and coaching chess conceptualization, I've used Chess.com ratings as my primary metric.

When people tell me their Lichess or their FIDE ratings, I translate those roughly to Chess.com ratings in my head so I know what I'm working with.

Over the years, I've built deep associations for those rating numbers.

1000 means one thing, 1200 another, 1500, 1800, 2000, whatever.

I know the big conceptualization challenges each of the major milestones faces.

When I started this series, I couldn't find a decent rating distribution graph for Chess.com, so I used Lichess's one instead, and to align with the graph, I decided to reference Lichess ratings throughout.

That was my first mistake.

Lichess ratings are roughly 2 to 300 points higher than the equivalent Chess.com ratings.

You can actually hear more about his background in episode three oh three back in two thousand twenty-two.

He's also an avid chess player, got into chess p-- through-- Well, he learned as a kid, but got into chess through his baseball connections, and he has been competing regularly and blogged about it on Lichess for a while.

And it's been four years since we chatted.

When we chatted, he said he had a long-term goal of earning a lower-tier title like Expert or Master, and at the time, he was rated about thirteen hundred USCF.

But w-- like, are you grinding like Chessable or Chessbook? You trying to memorize lines or you're playing through games? What is your actual approach to learning an opening? Especially like, you could walk us through a little bit the approach of switching from D4 to E4, because obviously that's a massive change.

Yeah.

Uh, I started, yeah, Chessable was such a huge part of my opening preparation and less so now, I would say just more than Lichess explorer and Chessbook and building out an opening book myself.

And what I think has been the most helpful that has allowed things to stick are trying to ev-- at least analyze three G... like strong master level games by hand per week.

But anyhow, we end up with the final position where we agreed to a draw.

Now I have it in Lichess.

Now this is an interesting thing.

Ours was the last game to finish.

And when we agreed to the draw, one of the players was like, you know, he's a higher rated player at the club.

He's like, "You know, maybe I, I, I think you're better here because your king was better." And I said, I go, "This is dead drawn.

I would bet money it's dead and drawn 'cause I'm pretty good in the endgame." So it turns out [laughs] I have it in Lichess, the evaluation where we agreed to the draw is literally zero point zero for, for like the first four lines in the engine.

So my assessment was correct, which I was happy about.

Is this an over-the-board tournament game? Is this an online tournament you're in?

in?I think it's Lichess

Ah.

Right.

Okay.

So you're in the...

Lichess is like the lone wolf league?

Lone wolf league.

That was enough to prototype the pipeline, but nowhere near enough to train our reliable model.

To increase both the size and diversity of the dataset, I combined three different sources: public games from Kaggle's Lichess dataset, games downloaded from professional chess, com accounts, personal games from my friend.

For the professional dataset, I selected players with very different profiles: Magnus Carlsen, Hikaru Nakamura, Ding Liren, and Levy Rozman, Gotham Chess.

Interestingly, Ding Liren only had four available games through the endpoint I used.

The core of the extraction pipeline simply iterates through each PGN game until the desired snapshot is reached.

API bottleneck.

My first idea was to retrieve games directly from the Lichess API.

Although it worked, it quickly became clear that downloading and processing thousands of games this way was far slower than expected.

All of the supporters there are helping the show go and keep going.

Uh, make sure you stop by our Lichess team tournaments.

I actually won this week.

They're every Sunday at 10:00 AM.