C
Evidence Grade C
Use with Hedging
Recent
3.9%
3.9% — hallucination rate for DeepSeek-V3 (2025)
Source:
View source
Curated by Aaron Agius
· Founder, Louder Online
Last verified:
·
How we verify data →
Source verified
Evidence graded
Value
3.9%
Unit
Percentage
Data Vintage
2025
Entity
Study Type
Unknown
Trend
Stable
Evidence Quality
C
Moderate quality evidence
Useful directional data with some methodological gaps. Treat as indicative rather than definitive — corroborate before citing in formal research.
Sources (2)
huxiu.com
Primary
Tier B
https://www.huxiu.com/article/4085480.html ↗
Source Article
本文来自微信公众号：阑夕，作者：阑夕，题图来自：AI生成
虽然DeepSeek-R1确实好用，但它在爆火之后，成了人手一个的AI工具，也对中文互联网的信息环境造成了严重的污染情况，这是一个固然难以避免但也理应得到重视的问题。
最近一个星期以来，就我看到的刷屏文章，至少有三例都是DeepSeek-R1生成出来的、充满了事实错误的内容，却因其以假乱真的迷惑性，让很多朋友信以为真，情绪激动地分享传播。
第一例，是知乎的这条高赞回答：
即使在我指出来之后，依然有人不可置信地表示，看不出来其中的“AI味”，所谓的“AI味”，指的是DeepSeek-R1创作文本时特有的“极繁主义”，比如生造概念、堆叠名词…
Also cited by
APPSO (via Huxiu)
https://www.huxiu.com/article/4122430.html
2026-05-03
Original Research
Vectara
· Original study ↗
Methodology
Vectara large model hallucination ranking
Study: Unknown
Cite this stat
Suggested hedging: based on claims without stated methodology, with unverified sample sizes...
Plain Text
DeepSeek-V3: 3.9% hallucination rate. Source: huxiu.com (2025). Via Lighthouse Research Data — https://lighthousedata.io/data/deepseek-v3-hallucination-rate-vectara
HTML Embed
<blockquote cite="https://lighthousedata.io/data/deepseek-v3-hallucination-rate-vectara" style="border-left:3px solid #2563eb;padding:12px 16px;margin:16px 0;font-family:system-ui,sans-serif;background:#f0f4ff;"><strong>3.9%</strong> — hallucination rate for DeepSeek-V3<br><small>Source: huxiu.com (2025) · <a href="https://lighthousedata.io/data/deepseek-v3-hallucination-rate-vectara" target="_blank" rel="noopener">Lighthouse Data</a></small></blockquote>
Share
Related Statistics
Use this data in your work
Copy the citation, embed the widget, or explore thousands more verified marketing statistics — all free.