#预测市场 Seeing the data changes on Kalshi's prediction market, I suddenly recalled several major reversals I've witnessed over the years. At the beginning of the year, ChatGPT was leading with a 41% probability, while Gemini was only at 30%. Who would have thought that nine months later, the situation would flip so dramatically—Gemini soaring to 86%, and ChatGPT dropping to 8%. The trading volume of $14 million also indicates that this contest is far from a niche topic; real money is voting here.

This reminds me of the ICO boom in 2017. Back then, people were also betting on various prediction platforms, full of confidence, repeatedly adjusting their probabilities. But history shows us that the consensus among market participants often collapses suddenly at a certain point. The current betting game around AI models essentially reflects that the pace of technological iteration has outstripped most people's cognitive updates.

From the actual performance of the LM Arena rankings, Gemini has indeed surpassed in multiple dimensions over the past six months. But I have also seen too many cases where "certainty consensus" was broken at critical moments. An interesting aspect of prediction markets is that they record not truth, but a snapshot of the collective cognition of participants at the moment. When the results are revealed at the end of 2025, will that 86% probability turn out to be a precise prophecy, or a collective mistake mocked by history? No one can say for sure now. The only certainty is that the change in this number itself is already a valuable record.