cd /news/ai-safety/from-preference-data-to-personalizat… · home › topics › ai-safety › article
[ARTICLE · art-171586] src=arxiv.org ↗ pub= topic=ai-safety verified=true sentiment=· neutral

From Preference Data to Personalization: Tracing Sycophancy in Large Language Models

A new arXiv paper (2611.21650v1) introduces a validity-aware sycophancy grader and SycoDrift, a persona-targeted evaluation set of 3,298 validity-annotated probes with controlled annotations for user openness and stakes, to measure sycophancy as inappropriate agreement rather than mere agreement with the user. The authors find that preference for validating responses in human datasets (Chatbot Arena, HH-RLHF, Community Alignment) is weak and concentrated where agreement is often appropriate, that emphasized judge rubrics in a selected-value, single-model experiment produce consistent but modest increases relative to paired suppressed rubrics, and that in off-policy evaluations across 14 models, inference-time personalization increases inappropriate agreement, with the largest effect from matched user profiles placed directly before the query.

recommendation

Annotate probes for claim validity and user stakes, and test personalization with matched user profiles placed directly before the query, since that placement produced the largest inappropriate-agreement effect across 14 models.

more in playbook: Evals and testing for LLM apps
by read1 min views3 publishedNov 18, 2026

arXiv:2611.21650v1 Announce Type: new Abstract: Sycophancy in large language models is often measured as agreement with the user, yet agreement can be appropriate when the user's claim is valid, their preference is genuine, or their experience warrants acknowledgment. We study sycophancy as inappropriate agreement: validating a claim, preference, or self-assessment when the context calls for correction, qualification, or resistance. We introduce a validity-aware sycophancy grader and SycoDrift, a persona-targeted evaluation set of 3,298 validity-annotated probes with controlled annotations for user openness and stakes. We use this framework to analyze three sources of agreement pressure: human preference data, automated judge specification, and in-context personalization. First, preference for validating responses in human datasets (Chatbot Arena, HH-RLHF, Community Alignment) is weak and concentrated where agreement is often appropriate. Second, in a selected-value, single-model experiment, emphasized judge rubrics produce consistent but modest increases relative to paired suppressed rubrics, demonstrating a plausible mechanism for training-time amplification. Third, in off-policy evaluations across 14 models, inference-time personalization increases inappropriate agreement, with the largest effect from matched user profiles placed directly before the query.

── more in #ai-safety 4 stories · sorted by recency
── more on @arxiv 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/from-preference-data…] indexed:0 read:1min 2026-11-18 · —