New Research: LLM Agents Lack Time Awareness, Can't Estimate Hours Spent on Tasks
solyarisoftware · x · 2026-08-18
Maksym Andriushchenko (Oxford) published a blog post, "Are LLM agents time-aware?", studying whether LLM agents can predict wall-clock time of tasks and estimate time they've spent, evaluated across long-horizon benchmarks including ProgramBench, PaperBench, and DeepSWE. Joint work with MATS mentee M. Ofengenden.
Rohan Paul highlights the implication: AI coding agents can spend hours on a task without a calibrated sense of time passing, so long-horizon evaluations may need to measure duration-following directly rather than treating sustained task performance as evidence that an agent knows when to stop.
Related event: New Oxford Research Finds LLM Agents Largely Lack Time Awareness(9 posts)→
More from Research
- U. Tokyo's TwinX robot: one wire links two wheeled-legged bots for morphing and teamwork — LexiLove · 2026-10-17
- NVIDIA open-sources midtraining recipe: 500B tokens lift Nemotron agentic benchmark scores — denisyarats · 2026-10-17
- PySR v2 ships after two years of work: plugins, custom types, much faster — MilesCranmer · 2026-10-17
- AAAI 2027 workshop on trustworthy AI agent evolution lands Bengio, Wang, Du as speakers — berkeley_ai · 2026-10-17
- T2-z Optimizes Environment Latents at Test Time for Robust Robotic World Models — yining_hong · 2026-10-17
- Study: AI agents steer wealthier-seeming users toward pricier options across 13 models — niloofar_mire · 2026-10-17