R0056/2026-04-01/C005/SRC01/E01

Research	R0056 — RLHF Yes-Men Claims v2
Run	2026-04-01
Claim	C005
Source	SRC01
Evidence	SRC01-E01
Type	Reported

Primary evidence for C005

URL: See source scorecard

Extract

Wei et al. found synthetic data finetuning reduced sycophancy by 4.7 to 10.0 percent across PaLM model sizes. Published at ICLR 2024.

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.