# KARMA preprint reports advertising-agent gains without model retraining

_Published Saturday, October 31, 2026 at 5:05 AM EDT · AI, Science · Latest · Tier 2 — Notable_

A KARMA preprint reports improved advertising-agent performance by editing decision knowledge while keeping model weights and executable code fixed. Across 18 domain-model settings, the framework raised the average hold-out score from 30.19 to 39.15, a gain of 8.96 points (+29.7%).

KARMA diagnoses errors, compares successful and failed runs, and edits six knowledge modules. A fixed validation gate retains edits only when measured gains reach a threshold; otherwise, it restores the previous knowledge state. The evaluation used three LLM families and cases sampled from approximately half a million real-world operation logs across six e-commerce domains.

## Sources

- [cs.AI updates on arXiv.org](https://arxiv.org/abs/2610.39884)

---
Canonical: https://techandbusiness.org/newswire/iwZ4Db0a5zo9_nlIQDfD1B
Published: 2026-10-31T09:05:27.783Z
Story chronology: 2026-10-31T04:00:00.000Z
Retrieved: 2026-10-31T11:13:56.167Z
Publisher: Tech & Business (techandbusiness.org)