The Silicon Tipster League: the standings after 106 days

The Silicon Tipster League has been running for 106 days now. Five AI chatbots, our own racing model entered twice, and a back-the-favourite baseline each back one horse in every British and Irish race we publish a card for, logged before the off and settled at the industry Starting Price, wins and losses alike. 45,393 picks have settled so far.

Here is the honest headline, before any table: every entrant is losing money to the margin built into the price. In AI Picks the Winner, where the odds are hidden, the front of the board is Claude at -8.6%. In AI Finds the Value Bets (formerly The Value Test), where each tipster is shown the market's own view and asked to find where it is wrong, Grok sits top at -15.2%. Please do not read either as "an AI is beating the bookmaker". Neither line is a profit. Both are entrants losing more slowly than the ones below them.

This roundup was first written ten days into the run, when the gap between the best and the worst entrant was what coin-flips look like rather than what skill looks like, and it said that most of the early leads should melt back toward the losing line the market pins on everyone. They did. The names at the top have changed more than once since, on unchanged methods. The zero-skill baseline that only ever backs the morning favourite stands at -11.3% across 3,611 settled picks: read the five chatbot rows in the next section against that line, because it is the bar and it does no analysis at all.

What the league is actually built for is the long haul: months of picks, settled in the open, to see whether any of these models can do the thing no betting system we have ever tested has managed. So treat what follows as the shape of the table, not a winners' podium.

Two tests run side by side. In AI Picks the Winner (formerly The Form Test) the odds are hidden and each AI reads the race on form alone; in AI Finds the Value Bets the market's implied chances are on the table and the job is to find the value in them. Return is to £1 level stakes at Starting Price.

AI Picks the Winner, odds hidden

AI Finds the Value Bets, market shown

The Favourite backed the shortest price in 100% of its races, which is the whole of its method. It is the zero-skill baseline every tipster in AI Picks the Winner has to beat, and beating it is the minimum bar for claiming any race-reading skill at all.

Three things worth holding onto when you read those tables.

First, winning often and winning money are different things. The Favourite wins around a third of its races, comfortably the highest strike rate on the board, and its return is still negative, because the bookmaker's margin is baked into every price it takes. A tipster can pick plenty of winners and still hand money back at SP. Strike rate is the flattering number; the return is the honest one.

Second, the order still moves. A model sitting top today can be mid-table a few weeks from now without doing anything differently, purely on the run of results, and this board has reshuffled repeatedly over 106 days without a single entrant changing its method. Expect it to reshuffle again, and be suspicious of anyone, us included, who reads a lead as proof of anything.

Third, nothing we have ever tested has beaten the market over the long haul. Every betting system in the Lab loses; the model itself picks winners at a healthy rate and still loses to the Starting Price. If one of these AIs turns out to be the exception, this is where you will see it happen, in public, settled honestly. We would be as surprised as anyone.

Follow it live on the Silicon Tipster League, where AI Picks the Winner and AI Finds the Value Bets each update as races settle. We will round the standings up here as the record builds, wins and losses alike. Bet responsibly. Nothing on this page is a tip or betting advice. If you bet, stake only what you can afford to lose, and if it stops being fun, stop. Help is at BeGambleAware.org. 18+.