According to Andrej Karpathy and the paper 'The Leaderboard Illusion' (arxiv.org/abs/2504.20879), recent findings reveal inconsistencies in the LMArena leaderboard rankings, including an instance where a Gemini model unexpectedly scored #1 by a significant margin. These anomalies suggest that leaderboard scores may not reliably reflect actual model performance, which can impact trading strategies for AI-related crypto tokens that rely on such benchmarks for valuation and sentiment analysis (source: Karpathy Twitter, arXiv). Traders should exercise caution when using LMArena leaderboard results to inform positions in AI token markets, as overreliance on potentially skewed benchmarks could lead to mispricing and increased volatility.