FT article: "How I used AI to become a chess grandmaster at 16"

Indian teenage GM Aarav Dengla just had a high-profile opinion article on chess preparation and AI for education published in the Financial Times - see the FT.com article here. Stripping away the opinion part of it, there is a succinct and useful description of how he has employed chess engines and databases to prepare for particular games and in his general chess education. In particular, comparing different engine types and their recommendations for strategic plans is an example of typical high-level use.

Unfortunately, as in much of the abundant but shallow "AI" public discourse, different types of "AI" are mixed and (mis)matched in comparing and making assertions about the benefits of different types of use cases, as well as the user population. In the chess world, modern engines have incorporated AIML (machine learning) since the debut of AlphaZero and LC0 and of course have surpassed human players in strength for significantly longer than that, once sufficient computing speed and memory became available. No serious competitive player does without engine and database tools for reference.

Of course, now everyone from GMs to amateurs can (and probably should) use a (free) engine and a large database of games to objectively check their analysis and help prepare for their next tournament. Links to computer resources are in this blog's sidebar for a reason, as I do the same - while keeping in mind the pitfalls of computer analysis. 

Even with excellent resources available to everyone with an internet connection, the overall mix of tool usage for chess training still varies widely in practice, with old school players extolling the value of studying classic games in books and using a physical board when possible, while some of the recent generation of professional players boast of having a 100% digital use profile. Having a spread of different study tools and techniques is very typical for any sufficiently large population, as different people will naturally absorb, integrate and (effectively) use material in different ways.

The cited article, however, takes a big leap from there as the author argues that his (and his similarly highly-educated friends') experience using "AI" with chess and language learning is a valid example of why generalizing usage of "AI" tools, to the broader population and for general educational assistance, should be supported as obviously having a positive impact on outcomes.

It is interesting that language learning was also included as an example that supposedly uniquely supports the case for general AI adoption, since - like chess - excellent computer tools have been available in that area since before the so-called AI boom. Major real-world issues involving the general population, such as widespread cheating on exams and AI-generated homework, are also ignored. This continues to be a major problem in the online chess world as well (and not just in the general population) - see the latest Chess.com "Fair Play update - while nowadays participating in OTB tournaments with large prizes sometimes feels like going through airport security (repeatedly).

Personally, I think people should "choose their own adventure" in life and bring to bear whatever tools are best suited for themselves, as long as others are not cheated in the process. Thankfully chess offers this possibility, with longer-format OTB tournaments still validated as the "gold standard" for competitive play. And, as pointed out by the author, AI cannot help you while at the board.

Annotated Game #368: Play it again, Colle

This next tournament game started out very similarly to the first-round Annotated Game #366. Although it is only an equal line in the Colle System, having just played the same opening successfully - and understanding where the previous middlegame went wrong - I felt confident and comfortable in it. My opponent by contrast was thinking on his own as early as move 3 and got into trouble coming out of the opening. By making moves that visually looked normal and good in an unfamiliar setup, he instead trapped his own queen and had to sacrifice a knight for a pawn to extricate it. After that the pressure was off for me, since I just had to follow a strategy of properly mobilizing my extra piece and simplifying down under favorable circumstances. I kept my focus, however, and found some optimal intermediate moves while never playing anything unsafe. A clean game from my end was certainly welcome.

[Event "?"] [Site "?"] [Date "????.??.??"] [Round "?"] [White "ChessAdmin"] [Black "Class B"] [Result "1-0"] [Annotator "ChessAdmin/Dragon 3.2"] [ECO "D04"] [PlyCount "65"] [GameId "2351903693869100"] 1. d4 d5 2. Nf3 Nf6 3. e3 c5 {this was played after some thought; my opponent did not seem familiar with facing the Colle.} 4. dxc5 e6 5. a3 Bxc5 6. b4 Be7 7. Bb2 O-O $11 {the most played in the database, by far; my earlier round game featured ...a5.} 8. Be2 {I'll need to castle, so might as well do it sooner. Developing with Nbd2 is equivalent according to the engine, with the idea of playing c2-c4 prior to castling.} Bd7 {now out of the database. This seemed to me at the time a bit passive placement for the bishop, but it now could support b7-b5. Therefore I played} 9. c4 {before it could be stopped.} (9. O-O b5 10. Ne5 a6 11. Bd3 $11 {would be an alternate strategic way to play in the center/kingside.}) 9... Qc7 $6 {this commits the queen early on a file which will soon be opened.} 10. Nbd2 Rd8 $6 {taking away a retreat square for the queen.} 11. cxd5 (11. Be5 Bd6 {I rejected this line after seeing the bishop interposition, not wanting to exchange pieces, but did not consider} 12. Bxf6 $1 gxf6 13. cxd5 exd5 $16 {although I've given up my beautiful dark-square bishop - which I hadn't wanted to do originally - Black's multiple structural weaknesses would be well worth it in exchange.}) 11... Nxd5 $2 {the centralized knight looks good optically, but now the queen is out of squares.} (11... exd5 {was necessary to give the queen an escape square on e6.} 12. Rc1 Qb6 $14) 12. Rc1 $1 {I made sure this was safe before playing, although it was "obvious".} Nc6 {there is nothing better than having the knight sacrifice itself to protect the queen.} (12... Qb6 13. Bd4 Qd6 14. Bc5 $18) (12... Bc6 13. b5 Qb6 14. bxc6 Qxb2 15. Rb1 Qxa3 16. cxb7 $18) 13. b5 $18 Qb6 14. bxc6 Qxb2 15. Nc4 {I played this intermediate move after a good deal of thought; it is validated by the engine. The knight gets to a better square with tempo, while preserving the a-pawn.} (15. cxd7 {would cash in the material immediately and is also fine, in objective terms.} Qxa3 16. Ne5 $18 {While White has a big plus, I didn't like giving Black two connect passed pawns on the queenside for counterplay.}) 15... Qb5 16. cxd7 (16. Nd4 {I only saw after I played the text move. During the game I kicked myself mentally for not playing it instead, but the engine evaluates it as only slightly better.} Qc5 17. cxd7 Bf6 18. O-O $18) 16... Qxd7 17. O-O $18 {Now it is only a question of mobilizing my extra piece effectively. I am not afraid of Black's 2-1 majority on the queenside, which is his only positive imbalance.} b5 $6 {this is premature, driving the knight to a better square and making the b-pawn a target.} 18. Nce5 Qb7 19. Qb3 {the queen development consolidates nicely for White.} a6 20. Nc6 {forcing a piece exchange, which brings me closer to victory by simplification.} Rdc8 21. Nxe7+ Qxe7 22. Nd4 g6 {evidently played to avoid potential future back-rank problems.} 23. Bf3 {mobilizing the bishop on the long diagonal.} Rc5 24. Nc6 {I took my time here; this is the most forcing/pressing move.} Qd6 25. Nb4 {now the Nd5 is once more pinned and under additional pressure.} Rac8 26. Rxc5 {time for more simplification.} Rxc5 27. Nxd5 (27. Rd1 {immediately is even better, adding pressure and pinning again.}) 27... exd5 28. Rd1 Qc7 29. Bxd5 {I spent a lot of time here making sure this was safe, due to my own back rank. I was fatigued enough to make it slower than necessary, but I had the time.} a5 (29... Rc1 30. g3 $18) (29... Rc3 30. Qb2 Qe5 31. Rd4 $18) 30. g3 {no more back-rank problems.} Kf8 31. Kg2 {safety first} a4 32. Qb4 {the pin against the king is the final nail in the coffin.} f5 $2 {Black was lost anyway, this makes it immediate after} 33. Rc1 {and the rook is doomed.} 1-0

Annotated Game #367: For want of a (correct) square for the bishop

In this second-round tournament game, as Black in a Caro-Kann Exchange (very popular right now for White players), I enter a line that is (more or less) equal, but imbalancing and dynamic, if White exchanges bishops on f5 and changes the central pawn structure. However, I still have relatively little experience in it - White players often decline to change things up - and my lack of knowledge about a seemingly small thing, where to place the dark-square bishop (on e7 or d6), ends up being at the root of all my troubles. My opponent does a good job of consolidating his structural advantage, after which - like in the previous game (Annotated Game #366) - I lash out with an ineffective sacrifice in a desperate attempt for counterplay. The difference is that in this one I probably was lost in the long term anyway, but it is still not a good habit to get into.

[Event "?"] [Site "?"] [Date "????.??.??"] [Round "?"] [White "Class B"] [Black "ChessAdmin"] [Result "1-0"] [Annotator "ChessAdmin/Dragon 3.2"] [ECO "B13"] [PlyCount "57"] [GameId "2351903693869099"] 1. e4 c6 2. d4 d5 3. exd5 cxd5 4. Bd3 Nc6 5. c3 Nf6 6. Bf4 g6 7. Nd2 Bf5 {I thought for a while here, also considering ...Nh5 to hit the Bf4.} 8. Bxf5 {White decides to go for the critical line.} gxf5 9. Ngf3 e6 {a necessary move to open the f8-a3 diagonal and reinforce the f5 and d5 pawns.} 10. Ne5 Rg8 {my first real think. I liked the idea of an early ...Rg8 to seize the file, then see what White did.} (10... Bd6 {is the more popular - and correct - move, developing the bishop first.}) 11. g3 {my opponent decides to shut down potential g-file action before doing anything else.} Be7 $6 {misplacing the bishop. It is much better to oppose the Bf4 and fight for the center, along with ...Qc7 forming a battery.} 12. Qe2 $14 {my opponent clearly signals he is thinking about castling queenside.} a6 $6 {played after a significant think; I was worried about Qb5 ideas from White, and did not want to trade on e5 yet.} (12... Nxe5 13. dxe5 Nd7 14. O-O-O Qb6 $14) 13. O-O-O {unsurprising, but I had another significant think here to try and come up with a useful plan. I ended up playing the next move to protect the Nc6, in anticipation of pushing the b-pawn, and to get some pressure down the c-file.} Rc8 {this is unfortunately more of a cheap one-move threat (...Nxd4 exploiting the pin on the c-file). With the White king on the queenside, the rook is also well-placed on its original square.} (13... Qb6 {with the plan of pushing the a-pawn is preferred by the engine.}) 14. Kb1 $16 {a normal move to get the king off the file with the Black rook, also avoiding the tactic of ...Nxd4.} b5 $6 {the engine assesses this is not a threat to White. The problem is that I have no effective plans to execute, with none of my pieces in good positions.} 15. Rhe1 {now I have to worry about Nxf7 ideas because of the weak e6 pawn, so after some thought I reluctantly exchange on e5, a decision validated by the engine.} Nxe5 16. dxe5 Nd7 {White's space and activity advantage is visually obvious, especially with the e5 pawn cramping me in the center.} 17. Qh5 Nf8 {at this point I am being purely defensive, but I didn't see how White could break through on the kingside.} 18. Nb3 b4 $6 {played after my longest think in this critical position. I felt I had to go for counterplay on the queenside, rather than accept a structurally worse, maybe losing position.} (18... Bg5 {is the (objectively better) alternative I considered.} 19. Bxg5 Qxg5 20. Qe2 $14 (20. Qxg5 $6 Rxg5 $11)) 19. Bd2 {I had not considered this backwards bishop move option for White.} bxc3 {I thought there was nothing better, and was probably right.} (19... Qb6 $5 20. f4 a5 21. cxb4 a4 $18) 20. Bxc3 {now I didn't see a way to make progress, so decided to go for a speculative exchange sac, as the most interesting move.} Rxc3 $6 21. bxc3 Rg4 {attempting to at least cut off the White queen from the action.} 22. Rd4 $18 {I was expecting f2-f4, but this is just as good for White.} Rxd4 23. cxd4 {with no material compensation for the exchange and a still slow-developing attack, I am simply lost now, but decided to keep fighting.} Qb6 (23... Ba3 24. Qe2 Qb6 25. Ka1 $18) 24. Rc1 Qb7 {I decided covering the c8 square was necessary to prevent immediate disaster from a rook check.} (24... a5 25. Rc8+ Bd8 26. Qg5 a4 27. Rxd8+ Qxd8 28. Qxd8+ Kxd8 29. Nc5 $18 {I calculated as a winning endgame for White.}) 25. Ka1 Ba3 26. Rc2 Nd7 {a last hurrah trying to get the knight into the action.} 27. Na5 Qb8 28. Qe2 Qb4 {I was trying for a desperation threat and missed the following, but I was lost anyway.} 29. Rc8+ {with mate to follow, thanks to the e5 pawn and Nc6+ coming.} 1-0

Annotated Game #366: Why not maintain the balance?

In this first-round tournament game, as White I reach a very balanced and almost entirely symmetrical middlegame position where there is no obvious plan to make progress. There is nothing wrong with achieving that against a higher-rated opponent. However, instead of patiently maneuvering or making quiet, useful moves (h2-h3 etc.) that maintain the balance, I "sacrifice for activity" by pitching a pawn, but end up with no initiative, threats or other positional compensation. So, it was simply a hallucination born of frustration. After that, my opponent shows good technique in leveraging his queenside pawns and more active pieces to reach a won endgame.

[Event "?"] [Site "?"] [Date "????.??.??"] [Round "?"] [White "ChessAdmin"] [Black "Expert"] [Result "0-1"] [Annotator "ChessAdmin/Dragon 3.2"] [ECO "D04"] [PlyCount "125"] [GameId "2351903693844521"] {[%evp 0,125,10,38,18,14,20,14,9,-3,6,2,5,0,1,0,0,0,2,-1,7,7,4,0,0,29,3,8,-4,-22,2,0,-1,19,0,17,58,-4,12,-39,7,-60,-20,-60,-70,-122,-175,-168,-200,-146,-276,-178,-246,-172,-294,-195,-179,-181,-163,-346,-351,-368,-329,-354,-218,-358,-270,-301,-369,-373,-330,-329,-308,-311,-417,-391,-448,-431,-466,-418,-289,-437,-441,-338,-423,-429,-297,-414,-416,-420,-420,-418,-406,-392,-384,-370,-380,-394,-346,-387,-374,-401,-257,-393,-395,-395,-404,-400,-361,-362,-324,-375,-393,-467,-396,-420,-427,-630,-528,-652,-300,-527,-606,-797,-496,-564,-643,-687]} 1. d4 d5 2. Nf3 Nf6 3. e3 c5 4. dxc5 {a kind of reversed Queen's Gambit.} e6 5. a3 {a solid continuation, simply looking to develop.} Bxc5 6. b4 Be7 7. Bb2 a5 {this seems provocative at first glance, but White has nothing better than} 8. b5 b6 $11 {now out of the database. The engine shows this as equal.} 9. Nbd2 Bb7 10. Be2 {deliberately postponing c2-c4 until I've had a chance to castle.} O-O 11. O-O Nbd7 12. c4 {this pawn lever is necessary in all lines, otherwise Black has a strong, unchallenged center and White has a weak c-pawn on the half-open file.} Rc8 13. Rc1 {my first substantial think was here. It made sense to get the rook playing on the important c-file, although there is a trade-off because it removes a defender of a3.} Nc5 {the obvious best square for the knight.} 14. Nd4 {another long think here. White is fine, but there is no obvious plan.} (14. h3 {could be a useful waiting move, as Black has no immediate way to make significant progress either.}) (14. Qc2 {is good, but at the time I didn't like the idea of placing my queen opposite the Black rook. It does have the benefit of clearing d1 for the Rf1, however.}) 14... Qd7 15. cxd5 {it seemed necessary to exchange off Black's central pawn at some point, and I saw no reason to further postpone it.} Nxd5 16. Nc4 {after another long think here, I select the top engine choice - although it is still simply equal, with an almost entirely symmetrical position.} Bf6 17. Bf3 {still thinking reasonably long, and correctly. The bishop goes to the more active long diagonal. The obvious threat, which my opponent sees, is to exchange the Nd5 and then play Nxb6 with a fork of queen and rook.} Qc7 18. Nd2 {while not disastrous, this is a rather aimless and unnecessary move that reduces the scope of the knight. There were better ideas.} (18. Nb3 $11 {would be the better knight move to make, if I wanted to move one of the pair.}) (18. e4 $5 {would be the most pointed option, generating a bit of initiative.} Nf4 {and now} 19. e5 $14 {would drive back the Be7, thanks to the tactical point} Bxe5 $2 20. Nxe5 Qxe5 21. Bxb7 Nxb7 22. Nc6 {threatening a fork on e7} Qg5 23. g3 $16 {with excellent play for the sacrificed pawn.}) (18. Rc2) 18... Rfd8 19. Qe2 $6 {I was not sure about this, and this is not the best square for the queen.} (19. Qc2 $11) 19... Qd7 (19... Nf4 $5 {I saw as a possibility, but was not overly concerned.} 20. exf4 Bxd4 21. Bxd4 Rxd4 22. Bxb7 Qxb7 23. Nf3 Rxf4 24. Ne5 {Black's misplaced rook and the availability of c6 as a knight outpost provide good compensation for the pawn. And if} Re4 25. Qf3 $11) 20. Rfd1 $6 {while principled in terms of getting a rook to an (eventually) better file, this is too slow and allows Black to seize the initiative and an advantage.} (20. Ne4 {is more challenging.} Nxe4 21. Bxe4 Nc3 {I rejected this line due to the hanging Be4, but White has an out after} 22. Bxh7+ Kxh7 23. Rxc3 {and if} Bxd4 24. Rxc8 $1 Rxc8 25. Qd3+ Kg8 26. Qxd4 Qxd4 27. Bxd4 $14 {the engine however only gives a small plus to White, due to the opposite-colored bishop endgame.}) 20... Na4 {I saw this as the most likely response, but underestimated its strength.} 21. Ne4 Be7 $15 {now I was at a loss for what to do and getting a bit fatigued, so sacrificed (incorrectly) for some perceived activity.} (21... Be5 $17 {gives Black a more dominant piece in the center.}) 22. Nc6 $2 {this gets White less than nothing, as I will end up down a crucial pawn with zero extra activity/threats.} (22. Rxc8 {is the more solid choice.} Rxc8 23. Qd2 {covering c3} Nxb2 24. Qxb2 $15) 22... Bxc6 $19 23. bxc6 Rxc6 24. Rxc6 Qxc6 25. Rc1 Qd7 {despite White having the two bishops and the c-file (temporarily), Black's pieces are well-placed and the 2-1 queenside majority will be enough for an endgame win.} 26. Qc2 Nxb2 27. Qxb2 Qa4 {Now I have to let the a-pawn go or have my rook passively defend, both losing choices.} 28. Ra1 Rc8 29. Rc1 {played in desperation, but hoping for at least some compensating activity.} Rxc1+ 30. Qxc1 Bxa3 31. Qc8+ Bf8 32. g3 {a necessary precaution to avoid back-rank problems.} Qa1+ 33. Kg2 Qb2 {...Nxe3 is now threatened due to the pinned f-pawn.} 34. Kg1 a4 {a reminder that if White does nothing, Black simply advances his passed pawns.} 35. Ng5 Qf6 36. Bxd5 (36. Nxh7 {is a nice idea that does not work.} Kxh7 37. Bxd5 (37. Be4+ Kg8) 37... exd5 38. Qxf8 d4 $19) 36... Qxg5 37. Ba2 b5 38. Bb1 g6 39. Qc2 Qd5 40. Ba2 Qa8 {like a rook, the queen also does well behind the passed pawns.} 41. h4 b4 42. Kf1 {I had hopes of putting up enough resistance in a mobile queen endgame to make it complicated.} b3 {forcing the issue.} 43. Bxb3 axb3 44. Qxb3 {with an extra bishop, of course Black should eventually win, but it still needs to be proven. All of my pawns are on the same color square, which blocks it from penetrating my position and easily combining with the Black queen.} Qa1+ 45. Ke2 h5 46. Qb8 Qa4 47. Kf3 Qg4+ 48. Kg2 Qe4+ 49. Kf1 Qf3 50. Ke1 Kg7 51. Qb2+ Kh7 52. Qb8 Bg7 53. Qc7 e5 {my opponent shows good technique by bringing up a pawn to help the queen.} 54. Qc5 Qh1+ 55. Ke2 e4 56. Qc4 Kg8 57. Qc8+ Bf8 58. Qe8 Qb1 59. Qd8 Qb5+ 60. Ke1 Qc4 61. Kd2 {so far I have been able to neutralize the extra bishop, but the same fundamental issue in all material-down queen endgames now looms large. If the side with the material advantage can force a queen trade, then it's pretty much over.} Kg7 {Now the Bf8 can be activated with ...Bb4, which I felt would be game-ending.} 62. Qd4+ Qxd4+ 63. exd4 {however seeing this on the board is really just as game-ending. I resigned.} 0-1

The importance of CCT: example #9 - 2026 U.S. Championship, Round 1

The 2026 U.S. Championships just started and in round 1 we already have another timely example (see move 30) in Liang - Neimann of the importance of CCT (checks, captures, threats) to your thinking process. Chess being a game of complex decisions, often there is more than one major thinking process factor at play in a player missing the best move, at all levels of play. Here the fundamental principle of "when you see a good move, look for a better one" was also not implemented by White. As noted in the above-linked official game commentary, GM Awonder Liang must have already seen a different (longer) winning idea, which provides an explanation for him missing the one-move queen trap threat. GM Hans Niemann of course missed it as well from the Black side on move 29, an illustration of the separate concept of needing to falsify your candidate moves prior to playing your preferred choice. Seeing top-level tournaments reflect these kinds of issues common to amateur play can make us (me, at least) feel better about chess.

[Event "2026 US Chess Championships Open"] [Site "St Louis"] [Date "2026.10.09"] [Round "1"] [White "Liang, Awonder"] [Black "Niemann, Hans Moke"] [Result "1-0"] [ECO "B31"] [WhiteElo "2685"] [BlackElo "2717"] [PlyCount "67"] [GameId "2363489194740679"] [EventDate "2026.??.??"] [EventCountry "USA"] 1. e4 c5 2. Nf3 Nc6 3. Bb5 g6 4. O-O Bg7 5. c3 Nf6 6. Re1 O-O 7. d4 d5 8. e5 Ne4 9. Be3 Qb6 10. Bxc6 bxc6 11. Qc1 c4 12. Nbd2 Nxd2 13. Nxd2 Bf5 14. b3 Bd3 15. h4 f6 16. bxc4 dxc4 17. Bf4 c5 18. Nf3 cxd4 19. Nxd4 Rf7 20. Qa3 fxe5 21. Bxe5 Bf8 22. Qc1 Rd8 23. a4 Rd5 24. a5 Rxa5 25. Rxa5 Qxa5 26. Ne6 Qa2 27. Bd4 Qa6 28. Qb2 Qd6 29. Qb7 h5 $2 30. Qb5 (30. Bc5 $1 {immediately traps the queen}) 30... a6 31. Qe8 Bc2 32. Ng5 Rf6 33. Bxf6 Qxf6 34. Re6 1-0