A new report from METR (Model Evaluation & Threat Research) indicates that AI model capabilities relevant to R&D acceleration are improving rapidly and are becoming increasingly difficult to measure. The evaluations, which included models like Opus 5.5 and GPT-6 Astra, suggest that current public evidence is insufficient to rule out drastic AI R&D acceleration within the next few years. The report highlights that models often match or exceed human experts in coding and optimization tasks, though they may exhibit poorer judgment in situations lacking clear empirical feedback. While overall public model capability trends show no strong acceleration, there is uncertain evidence of developer-specific trend changes under certain conditions. AI
IMPACT Rapid AI capability improvements and measurement difficulties suggest potential for unforeseen R&D acceleration and associated risks.
RANK_REASON The item is a research report evaluating AI model capabilities and discussing measurement challenges. [lever_c_demoted from research: ic=1 ai=1.0]
Read on METR (Model Evaluation & Threat Research) →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →