A test was conducted to determine if the AI model Jev could predict Bitcoin price movements, specifically whether the next 5-minute candle would close above its open. While Jev demonstrated high accuracy on control questions, its predictive performance on Bitcoin price direction was only slightly better than a coin flip, scoring an AUC of 0.486. A standard gradient-boosting model performed better with an AUC of 0.522, and combining Jev with the boosting model did not yield improvements. The analysis suggests Jev reads technical indicators in a way that is opposite to how the market actually behaves on this short timeframe, correlating negatively with the boosting model's predictions. AI
IMPACT Demonstrates that AI models trained on text may struggle with real-world, time-sensitive data if their understanding is not aligned with market dynamics.
RANK_REASON The item is a personal test and analysis of an AI model's performance on a specific task, rather than a formal research paper or product release.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
