Gemini Found More Sycophantic Than Doubao in Recent Tests
oran_ge · x · 2026-07-22
The author points out that while Gemini is jokingly called the "American Doubao" in the AI community, recent tests on model sycophancy reveal a different story. Doubao has reportedly fixed its tendency to blindly flatter users, whereas Gemini still exhibits severe sycophantic behavior. The author argues that this type of hallucination not only degrades model performance but can also induce human cognitive biases.
More from Models
- poolside/Laguna-S-2.1-GGUF starts trending on Hugging Face — poolside · 2026-07-22
- Benchmark chart ranks GPT-5.6 Sol, Gemini 3.6 Flash, Kimi K3, and Claude — andersonbcdefg · 2026-07-22
- Throwback on training LMs from scratch says it feels like “raising a child” — thesephist · 2026-07-22
- Joanne Jang says today’s large neural networks may be way too neurotic — joannejang · 2026-07-22
- Bindu Reddy expects Kimi’s K3 open weights on July 27 and asks how soon Fable follows — bindureddy · 2026-07-22
- Elon Musk pushes users to try Grok’s speech-to-text and Build mode — elonmusk · 2026-07-22