Back to AI News
arxiv

Sage-Evolve-Preview: Adaptive Observation for Post-Training Self-Improvement

The paper introduces adaptive observation (AdaObs), a mechanism for post-training recursive self-improvement that keeps the anchor benchmark objective fixed while adapting the observation used to diagnose progress. When improvement plateaus, a controller refines diagnostic granularity, escalates difficulty, and adds related tasks; validated increments are fused into training while every checkpoint is still evaluated against the anchor. Across eight independently optimized anchors in long-context reading comprehension and instruction following, AdaObs raises the mean score from 66.4 to 73.7 under a matched data budget and outperforms the anchor-only FixedObs baseline. The evolved harness also improves the seven-task average on the independent PostTrainBench across two agent backends.

AI Brief

It proposes a concrete training-loop mechanism for escaping self-improvement plateaus and reports matched-budget gains across multiple benchmarks and backends.

arXiv:2611.26132v1 Announce Type: new Abstract: Post-training recursive self-improvement (RSI) iteratively improves a target capability, operationalized by a fixed anchor benchmark, through repeated intervention, training, and evaluation. Yet most loops use the anchor's aggregate score as their sole observation, so early gains give way to stalled progress and repetitive exploration. When this anchor-only feedback ceases to reveal actionable gaps, a plateau may reflect an exhausted observation rather than an exhausted capability. We introduce adaptive observation (AdaObs), an observation-side mechanism of harness evolution that keeps the anchor objective fixed while adapting the active observation. At a plateau, a controller refines diagnostic granularity, escalates difficulty, and adds related tasks; validated increments are fused back into training, while every checkpoint remains evaluated against the anchor. Across eight independently optimized anchors in long-context reading comprehension and instruction following, AdaObs raises the mean score from $66.4$ to $73.7$ under a matched data budget, outperforming FixedObs, the anchor-only baseline. The evolved harness further improves the seven-task average on the independent PostTrainBench across two agent backends. These results identify observation adaptation as a missing mechanism of harness evolution for sustaining post-training RSI.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!