Gary Marcus argues that recent panic over AI progress, fueled by METR's "time horizon" graph, is misplaced. He contends the graph's 50% success rate metric is a low bar and doesn't reflect reliable performance or general intelligence. Marcus points out that improvements may stem from incorporating symbolic tools rather than pure model scaling, and the graph's focus on short software development tasks doesn't equate to broad human capabilities. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Argues against overstating current AI capabilities, suggesting progress is less about scaling and more about tool integration.
RANK_REASON Opinion piece by a known commentator discussing AI progress and a specific benchmark graph.
