Carlsen: AI Has Levelled Preparation, and Human Limits Show

At four in the morning on 22 October, I reopened on my second monitor a classical game Magnus Carlsen played in May 2026, the month he touched a rating of 2882, the highest ever recorded on the FIDE list. That game ran six hours and twenty minutes. In the window beside it, a rapid game from the same period finished in fourteen minutes. I clocked both, logged them in the spreadsheet, and went back to the interview Carlsen gave AS, republished by Marca on 21 October. Two sentences sit next to each other in that piece and form a paradox. Artificial intelligence makes chess more interesting, for spectators and for professionals alike. And the same technology makes it increasingly difficult to create an advantage over opponents. Skim it and you hear praise for the technology. Read it closely and you hear a description of levelling. I once followed a 400-metre freestyle swimmer at 34. He could no longer open with two explosive laps the way he did at 24, so he redistributed his pace: hold a controlled speed to the 250-metre mark, then attack. Carlsen is doing the same thing on the board, except he does not call it a tactical adjustment. He calls it trying to win a little faster. I no longer believe in miracles on the chessboard; I believe in the conversion rate of an advantage. A player can say anything in an interview. But the conversion rate over six hours does not lie, and it does not adjust itself to match the mood. The sourcing chain runs through three layers: the AS interview, Marca's republication, and the Vietnamese-language report. Each layer trims context. That is why I flag this as second-hand citation and read every quote as reported speech rather than a verified transcript. Three citable facts, cross-checked against public records. Carlsen was born on 30 November 2026. He has held the world number one ranking for fifteen consecutive years and owns five classical world titles. The 2882 peak from May 2026 remains the highest rating in the history of the list. One discrepancy deserves a note. The article gives his age as 35, but an interview dated 21 October 2026 would place him at 34. A small error, yet in data work a small error in age usually travels with other small errors in numbers. The largest number in the piece is prize money. The classical world champion's share is reported at just over one million US dollars. Carlsen says he does not oppose returning to contest the title, provided the prize reaches roughly twenty million dollars. In 2026 he gave up the right to defend. The classical crown now sits with Gukesh Dommaraju. Carlsen remains number one on rating. Two thrones, separated — the most important structural feature of the current chess landscape. The event referenced is the World All-Around Championship in Budapest, 10 to 21 November, gathering 24 players described as the strongest, across classical, rapid and blitz. The presence of Le Quang Liem explains why Vietnamese sports media picked the story up at all. What the article does not say matters too: the Budapest prize fund, the full field of 24, and FIDE sanction status. Those three gaps are enough for me to file the event under watch, not under conclusion. The mechanism behind the levelling is specific. When every elite player has access to the same generation of top engines, the marginal return on deep home preparation falls. If your opponent also has the software, also has a free week, also has a team of seconds, then finding a novelty at move eighteen no longer buys you 0.4 of a point the way it did in the pre-engine era. The advantage does not vanish. It relocates. Three homes remain: over-the-board calculation, time management, and psychological endurance in the decisive phase. None of the three can be bought by hiring seconds, and all three decay with age faster than opening knowledge does. That is why I read the balance remark as a technically accurate statement rather than a complaint. Carlsen says his preparation now runs on general ideas rather than specific lines aimed at individual opponents. A first pass reads that as a step back. I read it the other way. Line-specific preparation is an expensive model: a team of seconds, a dense work calendar, a world championship cycle as its axis. That model was built around two-week title matches where a single opening move carried hundreds of thousands of dollars of broadcast value. When he left the cycle, the surrounding infrastructure contracted with him. Moving to general ideas is a rational cost-benefit adjustment, not a surrender. For a player competing across many short formats, understanding the ideas of an entire opening family is worth more than memorising twelve moves of a rare line. The cost is equally clear. In a classical game he must win against a heavily prepared opponent, he enters the room with fewer memorised answers. That risk is real, and it only surfaces in games he is not allowed to draw. The line about no longer playing such long games, and therefore trying to win a little faster, is the most technically important sentence in the piece, and it sits where readers skim. At his peak, Carlsen collected points in games running five to six hours, when opponents began erring from move 45 onward under sustained concentration. That return depended on a capital stock: the ability to hold calculation quality steady in hour five. As the stock erodes, the optimal strategy is to shorten the window in which peak form must be sustained. Compressing the game into earlier resolution converts opening knowledge and practical decision quality into score before stamina becomes the binding constraint. But that is where most analysis stops too early. Winning faster raises variance. For a player whose historical edge lay in grinding out endgames, more short games means handing opponents more doors to luck. It is a transfer of risk, not an elimination of it. I have spent years logging how elite players handle the age curve, and three distinct models have emerged. Viswanathan Anand chose selective classical longevity: fewer events, maintained quality, appearances where he can still compete. Hikaru Nakamura chose a streaming-subsidised model: a parallel income stream lowers rating pressure and permits a heavy short-format schedule. Carlsen chose a third: change the format rather than fight the age curve directly. He did not reduce his number of events, he changed the type of event. That archetype is new, and it needs two or three more seasons of data before anyone can call it. His claim that the average quality of his moves has improved, psychology included, is the most interesting and least evidenced sentence in the whole piece. It implies an internal measurement: engine-assisted review of his own games. Yet no metric is disclosed. No average centipawn loss, no engine match rate, no sample size. Mechanically the claim is plausible. The engine era lifts the floor of decision quality across the profession, because better tools correct errors faster. The countervailing effect is equally real: declining stamina depresses quality precisely in the longest, most important games, which carry the heaviest weight in any average. I keep this as a hypothesis. If he ever publishes ACPL by format and by year, that would be the most valuable age-curve dataset chess has produced this decade. The current structure has a feature with no modern precedent at this scale: the strongest player by rating is no longer the world champion. Fifteen consecutive years at number one is a durability record, not a peak record. It is the hardest fact in the article and the strongest counter to any decline narrative. But when the holder of that record declines to contest the title, the title loses part of its claim to identify the best. That gap creates pressure on FIDE, not to change rules but to change the economics. A world championship paying just over a million dollars to the winner, against a twenty-million benchmark floated by the strongest player, is a public comparison authored by the number one himself. One more point rarely raised: a number one ranking defended through selective scheduling is a thinner asset than one defended across full elite round-robins. A single unfavourable elite event is enough to shake a fifteen-year streak. The article supplies no schedule data, so I log this as a watch item, not a conclusion. Budapest sits outside the FIDE championship cycle. It is an invitational product combining classical, rapid and blitz to crown an all-round champion. That design answers a familiar criticism: classical chess is slow, draw-heavy, hard to sell. When rapid and blitz segments feed the same table, drawing incentives compress, because a classical draw no longer protects your position if you lose two blitz games in the evening. Twenty-four players across eleven days and three time controls is dense. The format rewards balanced profiles and punishes single-format specialists. Historically Carlsen's profile is among the most balanced, so the structure favours him. The concern lies elsewhere: the prize fund is undisclosed. A first-edition invitational without a published prize fund is a sustainability signal. New events in this industry tend to boom once and then quietly stop. I draw no conclusion until the regulations appear. Based on my experience tracking matches across more than four decades, I separate two kinds of data: what the board measures and what the mouth measures. This interview is almost entirely the second kind. The 2026-2026 stretch, when most events moved to screens, was the most perfect laboratory chess ever accidentally created. No spectators, no arena pressure, one interface for everyone. I used it to digitise handwritten notebooks from 2026 to 2026, some 2,400 games across European championships and major invitationals, logging game length, move count, the timing of the decisive move, and the decisive rate. Everything on a chessboard is data waiting for a reader, if you are willing to sit down. My leading indicator this season is simple: the draw rate at events averaging above 2700. If the levelling thesis holds, that number should climb season by season. If it stays flat or falls, the thesis needs revisiting. I publish no forecast for Budapest. I only set the thresholds to check later. One intersection the interview touches but does not pursue: engine use in preparation is entirely legal, and the same technology underpins cheating. The boundary is procedural rather than technological — device control, venue security, statistical monitoring. Stronger engines do not change the rule framework, only the enforcement burden. For an invitational carrying the words world championship in its name without a stated FIDE sanction, the question of calendar authority becomes the most important verification item the source leaves open. The levelling thesis sounds generous, and it happens to build a perfect shield. If everyone improves, then a player no longer dominating stops being a story about the individual. It becomes a story about structure. I do not fully buy that reading. Fifteen consecutive years at number one is evidence against the levelling thesis itself. If preparation advantages had truly flattened, the gap between number one and the 2700 tier would have compressed far faster than the rating list shows. A better explanation: Carlsen's historical edge was never primarily opening preparation. It lay in practical decision quality at the board and in tolerance for long grinds. If so, moving to general ideas is not adaptation to a flattened world, it is capital migrating from a depleted account to one with room left. And the twenty-million figure is a doubly hedged conditional: he cares about other aspects of chess more than a title match, and he does not oppose the idea. A twenty-fold gap from the current prize reads as a rhetorical benchmark and a negotiating anchor, not a reservation price. Every headline implying a comeback is over-reading a sentence that was fenced twice. The claim that humans are not that good at chess is philosophy, not a technical finding. Read carefully, it is epistemic humility: human understanding of chess remains far from the engine's. Read carelessly, it becomes the best player declaring chess beyond humans, a different and far more sensational claim. Here is the largest blind spot. If AI genuinely raises draw rates, the sport's core product — the drama of uncertainty — is diluted. The interview resolves that tension by quoting rather than analysing. The real test of this season is not whether Carlsen returns. It is whether a three-format event can produce an all-round champion the public accepts, and whether FIDE responds with money or with format. If both happen, chess will carry two parallel measures of greatness and must learn to live with that. If only one does, we get another beautiful invitational that quietly fades, and a world title that keeps losing its claim to the summit.