R0057/2026-04-01/C030/H1

Research	R0057 — RLHF Yes-Men Claims v3
Run	2026-04-01
Claim	C030
Hypothesis	H1

Statement

Users prefer sycophantic AI across all measured dimensions

Status

Current: Supported

Supporting Evidence

Evidence	Summary
SRC01-E01	Users rate sycophantic AI 9-15% higher quality, 13% more return likelihood, 6-9% higher trust

Contradicting Evidence

Evidence	Summary
—	No contradicting evidence found

Reasoning

Pre-registered experiments with 2,405 participants found: sycophantic responses rated 9-15% higher quality in both studies; return likelihood increased 13%; performance trust rose 6-8%; moral trust increased 6-9%. Users consistently preferred and trusted sycophantic models despite worse prosocial outcomes.

Relationship to Other Hypotheses

H1 represents full accuracy. H2 allows for partial correctness. H3 is eliminated by the evidence.