# Preprint reports reduced variation in AI agent task progress

_Published Tuesday, October 27, 2026 at 10:04 AM EDT · AI, Science · Latest · Tier 2 — Notable_

Researchers report that a new training signal reduces variation in AI agents' task progress from 132% to 91% on tau2-bench and from 90% to 44% on toolsandbox. Their preprint introduces Recursive Semantic Divergence, a measure of how independent runs of an agent disagree when starting from the same state.

The method uses natural language inference to detect logical inconsistencies and a neural network to estimate disagreement across future turns. Used as an additional reward during training, it matches supervised fine-tuning performance with lower variance without requiring ground-truth labels. The findings concern benchmark tests, rather than demonstrated reliability in deployed systems.

## Sources

- [cs.CL updates on arXiv.org](https://arxiv.org/abs/2610.33012)

---
Canonical: https://techandbusiness.org/newswire/Q7v2W2BSxAkP1V3hRmta3L
Published: 2026-10-27T14:04:52.321Z
Story chronology: 2026-10-27T04:00:00.000Z
Retrieved: 2026-10-27T16:37:09.604Z
Publisher: Tech & Business (techandbusiness.org)