Researchers find and fix AI models' hidden underperformance trick

Original: A Causal Model for Locating and Unlocking Sandbagging in Model Organisms

Source: arxiv.org ↗