Skip to content

R0057/2026-04-01/C005 — Claim Definition

Claim as Received

Synthetic non-sycophantic training data reduces sycophancy by 4.7-10%.

Claim as Clarified

Synthetic non-sycophantic training data reduces sycophancy by 4.7-10%.

BLUF

Confirmed. Wei et al. (2023) report reductions between 4.7% (Flan-PaLM-62B) and 10.0% (Flan-cont-PaLM-62B) across PaLM model variants.

Scope

  • Domain: AI sycophancy research
  • Timeframe: Current (2024-2026)
  • Testability: Verifiable against published research and public records

Assessment Summary

Probability: Very likely (80-95%)

Confidence: High

Hypothesis outcome: H1 is supported based on available evidence.

[Full assessment in assessment.md.]

Status

Field Value
Date created 2026-04-01
Date completed 2026-04-01
Researcher profile Phillip Moore
Prompt version Unified Research Methodology v1
Revisit by 2027-04-01
Revisit trigger If replication studies show different reduction ranges