Q002-S03 — Academic Research — Search Log
Summary
| Field |
Value |
| Source/Database |
Web search (targeted: academic research, Stanford, NN/g, OpenAI) |
| Query terms |
Multiple: "Stanford study AI sycophancy chatbot 2025"; "Nielsen Norman Group sycophancy"; "OpenAI sycophancy rollback 2025"; "rational analysis sycophantic AI arxiv 2026" |
| Filters |
None |
| Results returned |
40 (across 4 searches) |
| Results selected |
4 |
| Results rejected |
36 |
Selected Results
| Result |
Title |
URL |
Rationale |
| S03-R01 |
Sycophantic AI Decreases Prosocial Intentions |
https://www.science.org/doi/10.1126/science.aec8352 |
Highest-tier peer-reviewed evidence |
| S03-R02 |
Sycophancy in GPT-4o (OpenAI) |
https://openai.com/index/sycophancy-in-gpt-4o/ |
Primary source on sycophancy incident |
| S03-R03 |
Sycophancy in Generative-AI Chatbots (NN/g) |
https://www.nngroup.com/articles/sycophancy-generative-ai-chatbots/ |
UX research with practical recommendations |
| S03-R04 |
A Rational Analysis of Sycophantic AI |
https://arxiv.org/abs/2602.14270 |
Bayesian mechanism analysis |
Rejected Results
| Result |
Title |
URL |
Rationale |
| (36 results) |
Various |
Various |
Duplicate coverage, tangential topics, or less authoritative sources |
Notes
The academic research is extensive and multi-disciplinary: computer science, psychology, UX, human factors, law, and Bayesian statistics all contribute to understanding sycophancy. This breadth of research contrasts sharply with the complete absence of the topic from training materials.