MIT Technology ReviewT2·Models
72 pts
AI evaluation nonprofit METR has updated its widely cited time-horizon plot, adding longer and more open-ended tasks. The trend of frontier models doubling the length of tasks they can complete roughly every four months appears to hold, but models remain weaker on judgment-heavy work than on coding. METR cautions that open-ended evaluation is still far from settling whether, or how fast, automated AI R&D might arrive.