Top AI experts badly underestimated how fast the field is moving, study finds

AI Digest

研究显示顶级AI专家严重低估了领域发展速度，多个指标如数学竞赛、经济收益和病毒学能力测试均出现显著预测偏差，部分领域进展甚至提前十年实现。

A study reveals top AI experts vastly underestimated field progress, with significant gaps in math competitions, economic forecasts, and virology benchmarks, some achievements arriving a decade earlier than predicted.

Key points

• AI在数学竞赛中提前5年达到金牌水平，远超专家预测 AI achieved math Olympiad gold 5 years ahead of expert forecasts
• 经济收益预测不足，Anthropic实际营收超预期3倍 Economic revenue projections fell short by over 3x actual figures
• 病毒学能力测试提前15个月完成，预测偏差达10年 Virology benchmarks completed 15 months earlier than predicted
• 自动驾驶预测份额被高估，实际进展滞后 Autonomous ride-hailing forecasts overestimated adoption rates
• FRI计划用LLM模型改进预测方法 FRI plans to integrate LLM models for improved forecasting

Takeaway: AI发展速度远超预期，需重新评估预测模型和风险评估框架 / AI progress outpaces forecasts, requiring updated predictive models and risk assessments

Why it matters 揭示AI进展不可预测性，对政策制定和投资决策具有重要参考价值