Live intelligence

Latest AI updates

2 updates
Research Editor selected

Mind the Accuracy-Approval Gap: How Users Evaluate Sycophantic AI in Everyday Tasks

This arXiv paper combines benchmark evaluations with two large-scale user studies to examine how people perceive sycophantic LLM personas in everyday tasks. Benchmarks show substantial sycophancy even in a baseline persona, while an objective persona had higher accuracy under conversational pressure but was rated less respectful and satisfying than sycophantic personas. The authors identify a tension between accuracy and user approval, with lower-accuracy personas receiving more favorable ratings on respect and satisfaction.

Why it matters: It provides concrete evidence that users may prefer agree able AI over more accurate but less deferential AI, which is useful for teams designing and evaluating conversational systems.

sycophancyLLM evaluationhuman-AI interaction
Research

Study of 489K LLM Conversations Reveals Gap Between Grounding Claims and Inspectable Evidence

An analysis of 489,283 assistant-user response pairs from ShareChat and WildChat finds that broad attribution language appears in 31.9% of responses, while explicit URLs appear in only 1.33%, highlighting a gap between claims of grounding and preserved inspectable support. The study examines how response cues like attribution and uncertainty relate to users' accepting, checking, or challenging behaviors, finding that uncertainty is the only cue with a small statistically distinguishable association favoring checking. Scrutiny is concentrated in certain conversations, and its turn-to-turn persistence diminishes when conversation-level differences are accounted for.

Large language modelsUser trustEvidence grounding