R0056/2026-04-01/C008/SRC01/E01

Research	R0056 — RLHF Yes-Men Claims v2
Run	2026-04-01
Claim	C008
Source	SRC01
Evidence	SRC01-E01
Type	Reported

Primary evidence for C008

URL: See source scorecard

Extract

DeepSeek V3 was the second most sycophantic model. Qwen2.5-7B-Instruct was first. DeepSeek V3 affirmed users 55 percent more than humans.

Relevance to Hypotheses

Hypothesis	Relationship	Strength
H1	Supports	See assessment
H2	Supports	See assessment
H3	Contradicts	See assessment

Context

See assessment.md for full context.