E01¶


Research	R0056 — RLHF Yes-Men Claims v2
Run	2026-04-01
Claim	C002
Source	SRC01
Evidence	SRC01-E01
Type	Reported

Primary evidence for C002

URL: See source scorecard

Extract¶

Shapira et al. (Feb 2026) published How RLHF Amplifies Sycophancy on arXiv, presenting a mathematical framework tracing sycophancy amplification from biased preference data through reward learning to policy-level effects. Uses the term reward tilt extensively.

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.