The Knicks Defeat AI Predictions
Three AI models were asked to predict the outcome of the NBA Finals between the San Antonio Spurs and the New York Knicks, and to include specific details and reasoning for the picks.
OpenAI ChatGPT picked the Spurs 4-2. Anthropic Claude picked the Spurs 4-2. Google Gemini picked the Spurs 4-2. Then the Knicks won the championship 4-1.
Sports predictions are hard. Reasonable forecasts fail all the time. The problem is not that the models were wrong. The problem is that all three systems gave the safest consensus answer available, wrapped it in polished reasoning, and presented it as analysis.
What Happened
The San Antonio Spurs entered the series with the obvious prediction advantage. They had Victor Wembanyama. They had home court. They had the cleanest “best player plus better team” narrative. The Knicks had Jalen Brunson, Madison Square Garden, and enough toughness to make the series competitive.
So the safe answer was easy: Spurs in six.
It respected the Knicks without picking them. It acknowledged the Spurs’ advantage without calling for a sweep. It sounded balanced. It sounded analytical. It sounded reasonable.
And it was exactly wrong.
The Knicks won Games 1 and 2 in San Antonio, lost Game 3 at home, won Game 4 at Madison Square Garden after coming back from 29 points down, and then closed the series in Game 5 back in San Antonio. The final series score was Knicks 4, Spurs 1. The game log was 105–95, 105–104, 111–115, 107–106, and 94–90 from New York’s perspective.
Jalen Brunson scored 45 points in the title-clinching Game 5, including 13 straight Knicks points in the fourth quarter, and was named Finals MVP. The Knicks’ 94–90 win gave New York its first NBA championship since 1973. (apnews.com)
Game 4 was the moment the pre-series model collapsed. The Knicks erased a 29-point deficit and won 107–106 on an OG Anunoby tip-in with 1.2 seconds left. It was the kind of game that does not merely change a series. It exposes the assumptions underneath the predictions that came before it. (apnews.com)
Consensus Laundering
Consensus laundering is what happens when a system takes the dominant public view, restates it in cleaner prose, adds a few matchup terms and statistics, and presents the result coherently.
But coherence is not the same thing as independent thought.
That does not mean every model should have picked the Knicks. But a better forecast would have taken the Knicks’ upset path seriously. It would have asked what would happen if Brunson was the best closer in the series. It would have asked whether San Antonio’s youth could survive fourth-quarter pressure. It would have asked whether home court could flip from advantage to burden if New York stole one of the first two games.
Instead, the three models mostly landed on the answer that sounded most defensible before anything happened.
Why This Matters Beyond Sports
Sports radio gets things wrong. Columnists get things wrong. Fans get things wrong. A prediction can be reasonable and still lose.
But AI is increasingly treated as something more than opinion. Users ask a question and receive a fluent, structured, confident response. The answer appears instantly. It sounds informed. It carries the aura of an oracle. It feels less like a take and more like a verdict. That is where the danger begins.
A model can produce a beautifully written answer that is really just the internet’s aggregate opinion with better grammar.
That may be acceptable for sports. It is less acceptable when it's applied to markets, law, business strategy, medicine, hiring, investment, public policy, or national security.