The paper re‑examines the METR time‑horizon metric, which reports the human completion time for software tasks that an AI can solve with 50% probability, by applying splines and item‑response theory to 228 tasks across 26 AIs. The authors find that the relationship between human time and AI difficulty is not strictly linear; the fitted spline is flat between 2–30 minutes and nearly linear outside that range, implying that a tenfold increase from 3 to 30 minutes is easier than from 30 minutes to 5 hours. They provide improved point estimates and diagnostic plots to assess the construct validity of time horizons, recommending that new benchmarks consider these diagnostics.
By Drew T. Nguyen, William Fithian
arXiv:2503. 14499v4 Announce Type: replace Abstract: Despite rapid progress on AI benchmarks, the real-world meaning of benchmark performance remains unclear.
By Thomas Kwa, Ben West, Joel Becker, Amy Deng, Katharyn Garcia, Max Hasin, Sami Jawhar, Megan Kinniment, Nate Rush, Sydney Von Arx, Ryan Bloom, Thomas Broadley, Haoxing Du, Brian Goodrich, Nikola Jurkovic, Luke Harold Miles, Seraphina Nix, Tao Lin, Chris Painter, Neev Parikh, David Rein, Lucas Jun Koba Sato, Hjalmar Wijk, Daniel M. Ziegler, Elizabeth Barnes, Lawrence Chan
arXiv:2602. 07267v2 Announce Type: replace Abstract: Evaluating the real-world capabilities of AI systems requires grounding benchmark performance in human-interpretable measures of task difficulty.
By Fengyuan Liu, Jay Gala, Nilaksh, Dzmitry Bahdanau, Siva Reddy, Hugo Larochelle
arXiv:2606. 07157v1 Announce Type: new Abstract: Many efforts to ensure frontier AI models are safe rely on monitoring their chain-of-thought (CoT) reasoning.
By Dewi Gould, Francis Rhys Ward, Anders Cairns Woodruff, Rauno Arike, Josh Hills, Alex Serrano, Ida Caspary, Jason Ross Brown, Jo J. Jiao, Patrick Leask, Twm Stone, Ram Potham, Ionut Gabriel Stan, Harry Mayne, Simeon Hellsten, Shubhorup Biswas, Ariana Azarbal, William L. Anderson, Elle Najt, Ryan Greenblatt, Julian Stastny
AgentTime is a benchmark that tests whether AI agents can manage their own runtime by working for a specified duration, predicting their wall‑clock time, and estimating elapsed time afterward. It includes 222 tasks from 18 sources across coding, computer use, agentic work, and automated research, with duration‑following experiments that append a single instruction specifying how long to work. Results show significant variability in accuracy, with some agents deviating from requested runtimes by up to 2.9×, and many agents stopping prematurely even after completing a task.
By Michael Ofengenden, Maksym Andriushchenko
Task difficulty dictates an agent's likelihood of success, and estimating it without rollouts means forecasting this directly from a task description before executing costly simulations in stateful environments. Reliable estimates would therefore allow environment designers to calibrate evaluation benchmarks and construct progressive training curricula.