Yesterday Google and Meta released new models back-to-back, and in benchmark tests the scores differed by more than 200 points between evaluation institutions. It looks like a bloodbath. But the AI community has figured out how to game rankings over these past two years: first, secretly test dozens of internal versions, pick the one with the highest score, and submit that to the leaderboard. This kind of operation was already exposed clearly in a 2025 paper called “Leaderboard Illusions.” Sounds familiar? In the crypto space, the TVL, TPS, and daily active user leaderboard play the same game: once an indicator becomes a KPI, someone starts gaming it. What truly determines who can stay is never the benchmark run on the day of the launch, but whether people are still genuinely using it three months later. Whether it’s AI or a blockchain project—don’t just look at the rankings, look at retention. $BTC $ETH #Crypto #AI #Web3