@Ra1kshit Was just running some benchmarks myself, and it seems much better, yeah! These games are great 👀
English
Asbjørn Steinskog
3.1K posts
Asbjørn Steinskog
@Assios
Dev @TakeTakeTakeApp · mod & volunteer @lichess
Oslo, Norway Katılım Temmuz 2009
571 Takip Edilen1.4K Takipçiler
For all the chess nerds,
I made google's latest Gemini Model battle out against its older variant in a series of 3 games in my harness.
3.5 Flash absolutely smoked 3.0 Flash ( note that these are SOTA models and claude, gpt dont come close in chess)
Sharing game links here lichess.org/study/CqBxh6pb…
English
@thechessnerd @GoogleDeepMind I've uploaded the 110 games played from the initial board position to these Lichess studies:
1-64: lichess.org/study/IeMBEw7t
65-110: lichess.org/study/WarhMxrO
English
Hi Chess Friends,
Does anyone have the PGNs of the 210 AlphaZero games @GoogleDeepMind released?
Their link seems to be dead:
English
@anishgiri @Ra1kshit i don’t think I can talk about that without risking a horse head in my bed
English
English
So if the lichess arenas are to remain the same, what are the expected taketaketake tournaments going to be?!
English
@anishgiri @Ra1kshit yes that will come, but not set in stone yet. any preferred formats?
English
@Ra1kshit that for sure, but they certainly did tease potential tournaments on multiple occasions (e.g. CEO interview with Dina).
English
Telling it to not invent numbers doesn't help much, it doesn't know that it's inventing them. Telling it to check a specific source like TWIC is better, but the best would be to connect it directly to the data, so the LLM isn't the one trying to find the facts, just reason on top of them and present them.
English
@SuneBergHansen Interesting!
Can you e.g tell it to check all the relevant tournaments on theweekinchess.com ?
Something like that would make it much more reliable indeed.
English
A very interesting example on the problems of using AI, but not double checking facts.
Claude starts inventing different historical results in Candidates and WC tournaments, leading to wrong statistics.
In reality, Anand scored +5 =7 -2, not +3 =11 in the 2005 WC tournament!
GM Jacob Aagaard@GMJacobAagaard
You can always write whatever story you want and make the stats fit... I would say that 2007, where your player, Anand, drew 10 games, has also higher. The second highest in the modern era, only exceeded by Anand without your assistance... Of course he also made 11 draws in 2005, where you helped him, but there he did not win. The matches where you were running one of the teams were very drawish. After, they were less so. The reason for the high draw percentage this time had two causes I can see. Firstly, at least two players, Wei Yi and Bluebaum, were not in fighting mood at all. Both seems unable to beat anyone, but are also quite hard to beat. The second reason is that Sindarov early on left many of the his competitors so far behind they "went into holiday mode" (that's a quote). I spent a little bit of time asking Claude to collect some statistics. Make of them what you want. There is no doubt that 960 is more difficult and should create more decisive results. You can also make the goals in football bigger and there will be more goals. Does it become more interesting? I would assume not. I had hoped the Freestyle Tour would be a success, but it simply was not. I wanted to like it, but it just did not feel like chess at all. And I have not yet given up on chess. claude.ai/public/artifac…
English
@Assios Are these roles open for US-based remote work? I’d love to be part of your team. DM?
ibrahim.py@def__ibrahim__
I tried to make the computer vision models as robust as possible by adding occlusions in the synthetic dataset. It should handle these well, but the main bottleneck is inference speed. It’s all still a work in progress. Would love to hear your insights or work with you, since you have experience with optical AI.
English
English
@ananyamiglani @TakeTakeTakeApp @Assios Mostly at that time Lichess server was restarting, Bad luck my friend, TakeTakeTake should have the same notification on Lichess that server will be restarting in X minutes
English
Hey @TakeTakeTakeApp !!
I played here with 00:00:15 seconds left in the clock and it was mate in one.
Eventually it was shown that I lost in time.
Does lichess binding delays premoves or raise the ping?
Coz every game is important and losing mate in 1 fckus with mind.
English
@Chessiro_ @chesscom Weird to thank your kidnapper for releasing you, but ok
English
@Chessiro_ I feel you! chessom is quoted in several Norwegian media as saying «innovation and competition is good for chess» these days btw😂
English
@BrainyMarsupial @TakeTakeTakeApp Thanks! Hmm, full game review with comments on all moves should be in place within ~10 seconds. If it takes significantly longer than that there could be a bug with the analysis. Feel free to send the game id if it continues to be slow (long-press your username to copy it).
English
@TakeTakeTakeApp @Assios Really love these features. Only major flaw right now is the speed. Sometimes the match review feedback is very slow
English
Asbjørn Steinskog retweetledi
"The game review explains what white was trying to do. White is trying to set a trap for the queen; it threatens to trap the queen by bishop g5, but it doesn't work."
@Assios explains how our game review understands and explains the moves that are being played, like a human coach would.
English
Asbjørn Steinskog retweetledi
The end of chessmafia as we know it.
With lichess x taketaketake, EndgameAI and ChessMonitor, the future of the internet chess world seems bright again.♟️🔥
English
Asbjørn Steinskog retweetledi
Ideally, serious chess esport events should be properly scheduled, held on a platform that doesn't glitch and with proper human proctors.
Just my opinion.
Greg Mustreader@mustreader
MVL on the 2026 CALENDAR CLASH and Hikaru’s CRITICISM. The podcast with @Vachier_Lagrave drops tomorrow (March 19) on the Chess with Mustreader YouTube channel! #Chess @GMHikaru
English
English
Thinking of moving to Linux permanently... Really liking it so far.
English
@DavidHowellGM Haven’t seen the UK version, but based on the Norwegian: The game design is actually pretty terrible! It punishes you for being right. I’d rather watch a show where celebrities just play Avalon
English
I love Traitors & all social deduction games. But UK series 4 (+ Celebrity) made it clear a revamp is needed. The Faithfuls too concerned with self-preservation to even try and figure things out. Maybe cast people who actually want to be Faithful? Too easy for the traitors atm 😪
English