AI

Reading the Room: Foundations, Design, and Challenges of Normative Competence in LLMs

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 09/10/2026, 04:03:24 UTC.

Briefings

This article presents a multi-agent debate setting to study normative competence in LLMs, where agents must learn social norms without static pretraining. The study reveals that baseline agents fail to learn norms even when doing so improves accuracy, indicating a fundamental inability to discern socially enforced order. Experiments with normative modules show that norm-following is highly sensitive to both the style of the norm and the specific model powering it, suggesting a lack of generalizability. Additionally, agents exhibit an unselective attribution failure, indiscriminately copying idiosyncratic noise alongside enforced rules, persisting even when penalizing unnecessary behaviors. The work is the first to operationalize and evaluate normative competence in LLMs, demonstrating that current systems excel at mimicry but lack the capacity to discern socially enforced order.

Key points

• Baseline LLM agents fail to learn norms even when doing so would improve their accuracy, indicating a fundamental inability to discern socially enforced order.
• Norm-following is highly sensitive to both the style of the norm and the specific model powering it, suggesting a lack of generalizability across different architectures.

Context and limitations

The study uses synthetic norms in a debate setting to isolate normative competence from pretraining exposure, but the findings may not generalize to real-world communities with rapidly changing or arbitrary norms.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 09/10/2026, 04:14:31 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeifnrqoa47otjrtw7kj5ivejxb6b6pfscg7tmhqzq7cclbwje32nfa