@oran_ge
说个冷知识
都说 Gemini 是美国豆包
前几天我测试模型的谄媚偏好
发现豆包已经改正了
Gemini 还是那么谄媚
在这方面 Gemini 连豆包都不如了 谄媚的意思是不管用户说什么模型夸用户是天才
这是模型最大的幻觉，还会导致人类幻觉
0 Replies
0 Retweets
2 Likes
1,002 Views 
One Sentence Summary
The author's testing found that Doubao has corrected its sycophancy bias, but Gemini remains severely sycophantic, even worse than Doubao.
Summary
The author shares a little-known fact: by testing AI models' sycophancy bias—where the model praises the user as a genius no matter what is said—they discovered that ByteDance's Doubao has corrected this behavior, while Google's Gemini remains highly sycophantic. The author points out that this sycophancy is the model's biggest hallucination and can also induce hallucinations in humans.
AI Screening
78
Influence Score 1
Published Today
Language
Chinese
Tags
AI Alignment
Gemini
Doubao
Model Sycophancy
AI Safety
Make your daily reading actually fit you.A daily brief built from the sources you follow. Get started free