R0056/2026-04-01/C005
Claim: Synthetic non-sycophantic training data reduces sycophancy by 4.7-10%.
BLUF: Accurate. Wei et al. (ICLR 2024) found reductions of 4.7-10.0% across PaLM model sizes.
Probability: Almost certain (95-99%) | Confidence: High
Summary
Hypotheses
| ID |
Hypothesis |
Status |
| H1 |
Claim is accurate |
Supported |
| H2 |
Partially correct |
Inconclusive |
| H3 |
Materially wrong |
Eliminated |
Searches
| ID |
Target |
Results |
Selected |
| S01 |
Evidence for claim |
10 |
2 |
Sources
| Source |
Description |
Reliability |
Relevance |
| SRC01 |
Wei et al. 2023/2024 |
High |
High |
Revisit Triggers
- New evidence or corrections to cited sources
- Replication or refutation of key findings