Skip to main content

Alfredo Merlet

2 related articles

Subscribe to updates

researchsquare.com Oct 22, 2026
An agent's time horizon shrinks by half when every retry must succeed, and the gap has not closed across model generations
Frontier AI agent time horizons are usually computed from average success: the task length an agent completes with 50% probability, measured in human expert time. This analysis instead fits a reliability horizon from METR's public run-level records, using the probability that all four attempts on the same task succeed. Across 50 model-benchmark combinations the reliability horizon is always shorter than the capability horizon, with a geometric-mean ratio of about 2.4, and the gap did not narrow significantly between 2023 and early 2026.
Frontier AI agent time horizons are usually computed from average success: the task length an agent completes with 50% probability, measured in human expert time. This analysis instead fits a reliability horizon from METR's public run-level records, using the probability that all four attempts on the same task succeed. Across 50 model-benchmark combinations the reliability horizon is always shorter than the capability horizon, with a geometric-mean ratio of about 2.4, and the gap did not narrow significantly between 2023 and early 2026.
Instituto Superior de Artes y Ciencias de la Comunicación
Alfredo Merlet（First author, Corresponding author）
Reliability horizon pass^k metric METR time-horizon evaluations Agent evaluation
researchsquare.com Oct 22, 2026
Virtual-world physics engines cannot hold a graviton: the correction only becomes storable beyond lunar orbit, while AI low-precision formats keep the signal
At the current graviton mass bound, the static correction from a massive graviton is too small for the floating-point formats of virtual-world physics engines to store: it first becomes representable at about 5.4×10⁸ m in float64 and 1.25×10¹³ m in float32. Naive in-format evaluation manufactures phantom corrections up to 9×10¹⁶ times the true value, and a float32 orbit at 1 AU reports 834 km where the true effect is 8 m. The same physics behaves oppositely in AI formats: every format, including 8-bit FP8, preserves the gravitational-wave dispersion residual, with representational SNR from 9.7 to 2×10⁷.
At the current graviton mass bound, the static correction from a massive graviton is too small for the floating-point formats of virtual-world physics engines to store: it first becomes representable at about 5.4×10⁸ m in float64 and 1.25×10¹³ m in float32. Naive in-format evaluation manufactures phantom corrections up to 9×10¹⁶ times the true value, and a float32 orbit at 1 AU reports 834 km where the true effect is 8 m. The same physics behaves oppositely in AI formats: every format, including 8-bit FP8, preserves the gravitational-wave dispersion residual, with representational SNR from 9.7 to 2×10⁷.
Universidad del Alba
Alfredo Merlet（First author, Corresponding author）
Massive Graviton Dispersion Floating-Point Representability Physics Engine Simulation Low-Precision Deep Learning

You’re all caught up