Is the AI Better Than Just Backing the Favourite?

We put a control in the Silicon Tipster League from the start, because a leaderboard of five chatbots tells you which chatbot, not whether any of them is any good.

The control is as simple as it gets. It backs the favourite in every race. It reads nothing, thinks about nothing and has no opinion about any horse.

It is currently ahead of every chatbot in the league.

That is the most useful single fact the league has produced, and it is the reason we run a baseline at all. Without it, Gemini's -15.7% would look like a result. Against it, it looks like a small difference inside a shared problem.

The five, against doing nothing

The favourite baseline is in that table too, ranked alongside everything else rather than pinned anywhere. 5 of the chatbots in it are behind a strategy with no intelligence in it at all.

Some caution is due. These are 55,124 settled bets over 127 days, the gaps are a few percentage points, and we have watched this order shuffle before. Nobody should conclude from a sample this size that the entrant at the top of that table is a better analyst than the one at the foot of it.

The robust part is not the ordering. It is that the whole group, intelligent and unintelligent alike, is clustered together and below zero.

Why a baseline matters more than a leaderboard

Any set of competitors produces a winner. That is arithmetic, not evidence. If five people flip coins for a month, one of them finishes ahead, and nothing has been demonstrated.

A baseline turns a leaderboard into a test. It asks whether the winner beat something that required no skill, and if not, the ranking is describing noise.

This is also why we keep the favourite baseline in view whenever anyone quotes a league position at us, including when the position is flattering. We would rather publish a result that undercuts our own experiment than run one that cannot fail.

The honest summary after 127 days: five modern AI systems, given a full racecard daily, have not demonstrated an ability to beat backing the favourite. That is not a failure of the experiment. It is the finding.

Our own model is held against the same baseline, race by race, on the track record page.

A word on all of this

None of these pages is a tip, and none describes a way to win. They describe what betting costs, which is a different and more reliable subject. If your betting has stopped being fun, BeGambleAware has free, confidential help, and the National Gambling Helpline is on 0808 8020 133.

Common questions

Can AI beat backing the favourite?

Not on our evidence. All five chatbots sit behind a favourite-backing baseline over 55,124 settled bets.

Why include a baseline that has no intelligence?

Because a leaderboard always has a winner. A baseline is what tells you whether the winner beat anything worth beating.

Does the leader change?

It has, more than once. The gaps are small relative to the noise over 127 days.

Is backing the favourite a good strategy then?

No. It returns -9.00% in our long-run testing. It is a low bar, and most of the AI tipsters are under it.

Would you publish this if the AI had won?

We publish the standings daily whichever way they run, including when our own model is last, which it has been.