Becoming the worst best chess player in the world
Like most people, late last year I watched the Queen's Gambit. Like most people, I made a Chess.com account assuming I’d become the new Beth Harmon. Unfortunately, more than literally almost everyone, I sucked at chess.
At first, this was unsurprising: I only vaguely knew what the pieces did, and didn't know more complicated rules like castling. But I'm a smart guy (hi white male privilege). If I just brushed up on the rules and played a few games I'd be beating Bobby Fischer in no time.
So I did. In December I played about 500 games of 'Blitz' (fast) chess. That's around an hour a day - it was lockdown after all. What did I learn? I suck at chess! Fucking. Suck. I was not even bad for a beginner, I was one of the worst people in the world.
After 500 Blitz games my ELO rating (the number that shows how strong you are) was 328. Out of the regulars on Chess.com, 11 million players, I wasn’t even 5th percentile. AKA, 95% of players on Chess.com were better than me. Worse still, I wasn't improving, I was getting worse. I slumped to 328 on January 1st 2021, after my month of Blitz.
The cursed game that took me to 328. In the spirit of Never Resign, I should have challenged my opponent to find Nb4#. Only I couldn’t find Kd3.
Now, my ego is quite large. And quite fragile. And I'd genuinely tried. I'd fallen in love with chess, but chess didn't love me back: it was starting to prove that I wasn't so smart after all. Well, we can't have that. I decided to figure out how to get good.
Given the size of my aforementioned ego, I wouldn't be writing about this if I hadn't made some progress. After a bit of research, I started a 20-30 minute a day practice routine. In early September/October I played two Over The Board (physical) chess tournaments. After the second tournament, my Rapid rating was 1508, about the same as my recent online ratings.
So, what's that mean? It means I went from being in the bottom 5% to pushing the top 5% of players on Chess.com in around 9 months. More importantly for you, dear reader: I now understand how adults can meaningfully improve at chess.
OK, some caveats. First, I'm no Beth Harmon: 1500 is not going to impress strong chess players. It's empirically harder to improve the stronger you are, and I have no evidence my methods will work for other people. Second, I now play almost exclusively Rapid (fast) games, instead of Blitz (faster). These ratings can't be compared precisely and from my just-had-a-few-beers-and-am-sitting-on-the-bus/train/toilet Blitz games my Blitz rating seems to lag by as much as 200 points. Third, the way I did this probably isn't fun for most people.
But, I did improve meaningfully. And, it turns out, in chess, almost no-one improves. Only about 10% of players ever gain more than 100 points, and only about 1% of players gain more than 200 rating points given years. Adult chess improvement is especially controversial. Since almost every chess Grandmaster became strong mostly during childhood, most strong players never had to do it!
So, if you too are an adult who sucks at chess, and want to improve, here's how.
I wish I had a single graph. To my shame I periodically made a new account, illogically hoping for different results. In October I hit 1492 on Chess.com and 1508 ECF OTB. Recently I’ve switched to Lichess where the rating system isn’t easily comparable, and sunk to 1497 OTB..!
What we're training
If you want my exact routine, skip to the next section. But, if you want to design a routine that works for you, you'll need to understand why my routine worked.
Especially at lower levels, chess is a game of short term patterns, not long term strategy. Sure, you can make a plan. But, mostly, it won't matter. It turns out trying to win a chess game on strategy alone is like trying to win a boxing match without throwing punches.
Why? Because your chess games are almost always decided by patterns called 'tactics' - a short sequence of moves that turns your roughly equal game into a completely unequal game. If you've played chess, you'll know the feeling: you’ve blundered a tactic every time your opponents took your Queen for free. Tactics and blunders are two sides of the same coin: blunders are move that create tactical opportunities.
Thanks for the tip Chess.com Coach.
And if you can't spot tactics, you can't win games. You can play openings like Magnus Carlsen, but play endings like Carlos Magnussen and you’re still going to lose. In the Woodpecker Method, GMs Axel Smith & Hans Tikkanen estimated what percentage of games are won or lost by tactical mistakes at different ratings:
GrandMasters 42%
2200-2400s 44%
2000-2200s 63%
1800-2000s 72%
300-500s 100%*
*my estimate. GMs Axel and Hans didn’t review my games.
Sadly this doesn't mean GMs don't do tactics. It means tactical skill is chess’s foundational skill, which they’ve (modestly) transcended. This classic De La Maza article has more.
But what is tactical skill? Let's check the science.
Well, the simpler the tactic, the more likely a strong player is to consider the best move immediately. In a position with a tactical opportunity, this study found that the first move that pops into a 2500 players head is as strong as the move that a 1548 player takes 5 minutes to find. And, fascinatingly, both strong and weak players improve decision making at similar rates given more time.
So what? The stronger the player, the more automatic the recognition of tactical patterns. They don't see individual pieces, they see chunks. Strong players recognise tactics like you recognise faces, or words, or tastes. Capablanca's famous adage "I see only one move ahead, but always the best move" turns out to be true.
If fluency in a language is measured in accuracy, speed, and prosody, then stronger players are more tactically fluent in chess. They find better moves, right away, in more relevant positions. Which begs the question, if tactical fluency is your goal, how should you train?
To get accurate:
Test yourself. You'll learn more efficiently using Active Recall than Passive Recall.
To get fast:
Study ‘Comprehensible Input’. You need to be capable of independently deriving the patterns you’re testing. Start from the absolute basics. Krashen explains here.
Test response time. We process facial expressions and complex sentences in milliseconds - why not chess positions? Remember, we're trying to train your automatic pattern recognition and chunking, not your calculation or knowledge.
To see a pattern in more positions:
Learn by category. To learn the pattern of 'mate in one with a queen', test yourself on lots of them in series, with instructive commentary. Context and subtle differences build understanding, on top of raw memory.
Then test randomly. Use a collection of randomly ordered patterns to encourage your brain to spot familiar patterns in new and more complex positions. You're trying to generalise pattern recognition when you don’t know what you’re looking for. Here's more on why categories and randomness work.
To see more patterns:
Use Spaced Repetition. For any given test, you should repeat the same test with more and more space in between tests. This dramatically reduces study time. Gwern has the best write up on Active Recall and Spaced Repetition here.
Once you're accurate, fast, and can spot the pattern in almost any context, you're fluent in a tactic. The more tactics you're fluent in, the more games you'll win.
My Routine for Tactical Fluency
This is the routine and tooling I used to train tactical fluency. To understand why this works, check the previous section.
The simplest way to start
The simplest recommendation would be to train Chess Tempo ‘Blitz’ puzzles on Easy mode. Chess Tempo Blitz scores you on whether or not you can solve a puzzle, and how fast you solve it relative to the average solver. If you are faster than average, you get more points. Much slower, you might even lose points. Get it wrong, lose even more. The puzzles are generally well curated, with lots of human commentary.
Another alternative would be to solve Lichess puzzles on the ‘very easy’ setting. This will only give you puzzles that are simple to solve and can be solved quickly, and will penalise your puzzle rating severely for making a mistake. It works for pattern recognition, because even simple errors will make the next puzzles you see a lot easier.
Both these options are free or close to, and easy to get started.
My approach
Before you think this is an ad - I'm not affiliated with Chessable in any way, it's just perfect for what we need. Each tweak of the setting is explained so you can replicate this in a different system if you wish.
I’ve made some small improvements to this approach since this article was published. Check the ‘Update’ section at the end of this article for those tweaks.
First, software.
Chessable is a Spaced Repetition platform for chess. You do tactics, and if you get them right it shows you the tactic again in the future (a 'review'). You can even set time limits for answering. Chessable makes it easy to test accuracy and speed with spaced repetition. This is where we are going to spend all our time.
Second, courses.
We need a course which explains and groups tactical patterns, and a course which presents them randomly. This means we learn patterns and how to apply them across contexts.
For categorised patterns and explanations, I have found GM Susan Polgar's Learn Chess The Right Way series to be incredible, but there are probably others. I am on Book 4 right now, but I especially loved Book 3 on defensive tactics.
For random tactics, I found Tactics Time to be good. It is basic stuff, jumbled up, which is what we need.
Readers have recommended Common Chess Patterns for beginner / intermediate, and Improve Your Chess Tactics for intermediate / advanced players. Both contain categorised patterns and random tactics and are very well reviewed.
Third, Chessable settings.
I recommend studying tactics with some specific settings. These settings force us to solve many problems we can understand slowly, fast.
You'll need a Pro account, but it's great value. You'll only need to change these once.
Once you've got a course, set your course settings to:
Study: Key Moves. This means we don't get tested on long unforced ends of variations.
Review: Whole Variation. This means we always get tested on each tactic from start to finish, rather than each individual move separately. Ideally we play through the whole variation in our head before making our first move.
Reps: 1. This means if/when we get shown the answer to a tactic, we only get shown it once. We wan't to find the answer rather than drill it.
Depth: Full Depth. This means we don't stop the tactic if it is more than N number of moves long. We want to study tactics start to finish.
Soft fail: Retry on alternative good move. This means we don't fail a tactic because there's multiple ways to mate and we pick the 'wrong' one.
Time: 8 seconds. This means we can only get a tactic correct if we answer within 8 seconds. This is important - we don't just want to get the answer right, we want to get it right almost instantly. I've messed around with different numbers here and it seems like 8 hits the sweet spot of 'I have time to see the position', 'I have some time to visualise the tactic from my first intuited move to the end', and 'if my first intuited move is wrong I don't have time to think of another'. The right number here is unclear, but 8 seconds isn’t unreasonably quick. Puzzle Rush asks you to solve progressively harder puzzles (going far beyond our level) for 5 minutes and the top scores are sub 5 seconds per puzzle.
Tactics: Solve problem. This means we don't want to be shown the answer to a tactic in advance, we have to solve it (or fail it) first.
Schedule:
Level 1: 1 day 16 hours
Level 2: 4 days 16 hours
Level 3: 10 days 16 hours
Level 4: 24 days 16 hours
Level 5: 56 days 16 hours
Level 6: 128 days 16 hours
Level 7: 296 days 16 hours
Level 8: 681 days 16 hours
This is a much more aggressive schedule than Chessable's default, both beginning with a larger interval and increasing the space between each correct review by ~2.1.x. This is actually conservative based on my experience with Anki, which uses an even higher default interval growth multiplier of 2.5. There are two major benefits to aggressive spacing, based on the idea that your fluency is cumulative:
In the short term it encourages understanding, rather than (just) rote memorisation. It’s better to have longer intervals even if you fail more. We’re not trying to optimise for the probability of getting any specific puzzle correct, we’re trying to optimise for cumulative absorption of the pattern.
In the long term it reduces workload per tactic which increases the amount of tactics you can maintain. If you average 12 seconds a tactic (time slack given failures) at Level 6, you can maintain 12800 tactics in just 20 minutes a day! Maintaining these 498 checkmates now takes me 3-4 reviews a day or about 45 seconds, with an average interval of 142 days per tactic.
NB. 16 hour thing is because Chessable doesn't make everything due by 'day', and so we basically do it manually by assuming you'll review at a similar time each day and scheduling each tactic to be due before then (eg a tactic due in 1 day 16 hours is in practice due in 2 days). Without this managing how many reviews you have per day becomes impossible.
Finally, look for the 'MoveTrainer' Settings (a little gear icon when studying on web and in the app’s Settings on mobile):. Then change these settings:
Time up action: stop timer. This means when the 8 second timer runs out, even if you get the right answer you will still have failed the tactic... but, you will have unlimited time to find this right answer. This is important: if after 8 seconds you are just shown the solution, your brain will not have to understand the tactic and you may not be training Comprehensible Input.
Enable retry: on. This means that if you answer incorrectly, you will get another attempt to find the right answer. This is also important. While we are training our automatic pattern recognition, we want to force our brain to understand the tactic in cases where our intuition fails.
Max retries for a mistake: 1. If we generate a wrong answer, we want to force ourselves to properly solve the tactic and find the correct answer, rather than to allow ourselves to just try multiple incorrect answers until one passes.
Retry action: stop timer. As above, we want unlimited time to find the right answer if previously gave an incorrect answer, to aid understanding.
Time up action for retry: stop timer. As above. understanding.
Highlight legal moves: off. This means when you click or grab a piece, the moves it can legally make will be highlighted on the board. This is a crutch you won’t have OTB so turn it off.
However you do it, get started and keep going.
Set aside some amount of time you can afford every day, and then split it across studying tactics by category and then randomly.
For me, I set a 10 minute timer, and then review the Polgar categorised tactics courses, from easy to hard. If I finish my reviews within the 10 minutes, I learn more categorised tactics. After 10 minutes, I finish the problem I'm on.
Then, I set a 10 minute timer, and then review randomly ordered tactics from Tactics Time. If I finish my reviews within the 10 minutes, I learn more random tactics. After 9 minutes, I finish the problem I'm on.
That's it.
It takes around 20-30 minutes a day, depending on how I scale up or down the time. One last thing: Spaced Repetition punishes inconsistency because reviews clump up. Spend less time on Chessable than you can spare each day, in case you miss a day. If you have extra study time one day, don’t spend it on Chessable unless you can afford more review time in the coming days. Both under and overstudying create review backlogs, which are horrible to clear.
My Chessable review heatmap.
FAQs
What about openings?
I learned a bunch of openings with White for 6 months or so, also via Chessable. Amazingly I won more games with Black, where I had learned nothing, than with White. I got frustrated with this, and switched my openings entirely. It had literally no impact on my rating, and I continued to improve.
I did a quick calculation using the Lichess database and realised, outside GM level, the most common lines still occur less than 0.5% of the time after 8 moves, even assuming you remember your moves 100% of the time! Basically, memorising lines is pretty useless.
I do still do a few minutes a day on low depth openings, because it gives me psychological safety - "I know what I'm doing" - and it’s fun to try different things.
What about pawn structure? Or strategy?
I have no idea!
What next?
Once I've finished the Polgar series and Tactics Time, I plan to apply this method to more advanced courses, like:
Basic Endgames, The Checkmate Patterns Manual, Dazzling Defenses against Checkmate Patterns, The Woodpecker Method, Common Chess Patterns, Improve Your Chess Tactics
I can't vouch for these yet though.
Do you think you can become a titled player with this method?
Probably not. If chess fluency is like language fluency, there’s probably a reason most GMs started very young. Getting a title means beating native speakers.
Should I play games?
Games are obviously fun, but not obviously good for your improvement. I have averaged about one online Rapid game per day this year. OTB tournament games have a different feel and are more memorable, so I would prioritise those from a learning perspective.
What time control for games?
Early on I played almost entirely Blitz, but the prevailing wisdom from legendary coaches like Dan Heisman seems to be that longer games are better for learners. I now mainly play Rapid.
The risk of Blitz and Bullet is that you drill your bad intuitions without corrective feedback or understanding - literally the opposite of what we're trying to do! The issue with really long games is that you can't play many of them, and in most positions there's diminishing marginal returns on quality thinking time.
Playing without increment has a similar effect. We don't want to force ourselves into playing endgame Bullet because we thought about a position for a few minutes.
For me, Rapid with an increment strikes a good balance. Sadly, I am not immune to the charms of toilet Bullet.
Where to play and what to play with?
Chess.com and Lichess are both great sites.
That said, in my first OTB tournament, I found that having learned chess entirely on a 2D plane with 2D pieces meant that my tactical fluency was dramatically worse in real life. I started using Lichess, because it's available on mobile web with a good looking 3D chess board which matches the plastic OTB tournament sets (see pic).
I genuinely think the same 3D pieces would improve Chessable's utility a lot (if anyone from there is reading!).
And turn off Highlight Legal Moves. You don’t get that OTB.
What about other tactics and tools? Puzzle Rush? Tactics Trainer?
I don't know, really. While I think these could be good benchmarking tools, I don't know if they're comparatively good tools to train tactical fluency. They tend to encourage speed or understanding, rather than both together. Puzzle Rush doesn't have a way to promote understanding, and Tactics Trainer ends up showing you puzzles that are beyond your Comprehensible Input.
I do like to play Lichess/Chess.com tactics on my phone in the background while watching TV. I don't focus on it, and just allow myself to take as much time as I want to solve a puzzle - often a whole evening or not at all if the show is good. I am around 2200 on both apps right now.
What about Chess content?
I confess, I also love a Gotham Recap. And who doesn't love watching GM Simon Williams go 0-8 playing the Reverse Double Bird Sicilian Gambit into a 2200 player?
How did you come up with this stuff?
I transplanted a lot of methods from language learning. Some things I took from folks like Eshepard, JA-Dark, (the early days of) AJATT, and Heisig. For Chess specifically, I got a bunch of ideas about how to learn Chess from the Perpetual Chess Podcast by Ben Johnson. The episode with Logozar sharpened my approach further.
Haven’t you just spent a lot of time playing chess lately?
A lot for some people and a little for others. But how much I play doesn’t correlate with my rating, which is true for most people.
What did you do in January?
Tactics on Chess.com, and some Woodpecker Method style stuff which lead me to my current approach.
Can you share your online IDs?
I have never played chess during a meeting.
Is this a fun method?
Not even for me! But if you like improving, or winning, maybe you'll get over that.
The “plastic tournament” set on Lichess.
Updated method 1 year later
I’ve since tweaked some elements of this approach:
- increased the time to solve to 10 seconds, but hidden the solve timer from screen (easiest done by scrolling so the timer is off screen - better done by just removing the timer element with an ad blocker). The goal is to solve faster, not (just) to guess faster and feel time pressure.
- focused on fully visualising the critical line of the puzzle to it's end position. So, rather than playing the first best move before the time runs out, I have to be able to fully visualise the critical line to its end position. If it’s a mate this would include seeing the king has no escape squares, rather than just the mating move, for example.
- limited the books I’ve studied to exclude random tactics collections (eg tactics time), favouring collections focused on specific patterns (eg Learn Chess The Right Way). This ensures I’m initially learning a comprehensive group of patterns presented in a way I can easily absorb, and then the spaced repetition takes care of automating them. For a full list of books I’d recommend, that I am working my way through, go here.
- now only use use ‘review all’, rather than review structured tactics then mixed. Over time this jumbles the tactics together as I get throngs wrong, and as the due times overlap. You get the best of both: structured patterns as you’re learning and then interleaved mixed testing.
- increased the initial interval to as much as 7 or 30 days, scaling at 2 - 2.5 rate. You can find the full intervals here. 4 days is the data backed most efficient first interval on Supermemo, but I personally feel longer works for chess, since we’re not optimising for any given puzzle. This schedule allows you to get less bogged down by puzzles you find tough, while taking advantage of the cumulative effect of doing lots of puzzles and making pure memorisation harder.
Update 2 years later
This somehow become the top ranked result on Google for “how to learn chess”. Good news: it still works. I hit 2k on chess.com doing pretty much this, with a bit of calculation added in. OTB is another beast.
Thanks to Anne Marie Droste and Matt Sharp for proof reading. And thanks especially to the language learners, Logozar, Chessable, and Ben Johnson for their ideas and inspiraton.
How did chess follow me here 😭🙏
I used to play on chess.com, but there’s so much cheating! It’s very frustrating. I play only as a guest, and I don’t worry about my Elo.