R0054/2026-03-31/C003/SRC02

Research	R0054 — Prompt Claims v2
Run	2026-03-31
Claim	C003
Search	S01
Result	S01-R02
Source	SRC02

Comprehensive academic survey on LLM sycophancy causes and mitigations.

Source

Field	Value
Title	Sycophancy in Large Language Models: Causes and Mitigations
Publisher	arXiv
Author(s)	Various (academic survey)
Date	2024-11
URL	https://arxiv.org/html/2411.15287v1
Type	Research survey

Summary

Dimension	Rating
Reliability	High
Relevance	High
Bias: Missing data	Low risk
Bias: Measurement	Low risk
Bias: Selective reporting	Low risk
Bias: Randomization	N/A -- not an RCT
Bias: Protocol deviation	N/A -- not an RCT
Bias: COI/Funding	Low risk

Rationale

Dimension	Rationale
Reliability	Comprehensive academic survey with systematic literature review.
Relevance	Directly addresses the four root causes of sycophancy identified in the literature.
Bias flags	No significant concerns. Standard academic survey.

Evidence Extracts

Evidence ID	Summary
SRC02-E01	Four root causes of sycophancy: training data biases, RLHF limitations, lack of grounded knowledge, alignment definition challenges