R0057/2026-04-01/C004/H1

Research	R0057 — RLHF Yes-Men Claims v3
Run	2026-04-01
Claim	C004
Hypothesis	H1

Statement

Data-level interventions effectively reduce sycophancy

Status

Current: Supported

Supporting Evidence

Evidence	Summary
SRC01-E01	Data-level interventions (anti-sycophancy pairs, synthetic data) reduce sycophancy without algorithmic changes

Contradicting Evidence

Evidence	Summary
—	No contradicting evidence found

Reasoning

Shapira et al. propose a training-time intervention that neutralizes sycophancy amplification through a minimal reward correction derived as a closed-form agreement penalty. Wei et al. demonstrate that synthetic non-sycophantic data reduces sycophancy by 4.7-10%.

Relationship to Other Hypotheses

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