[爆卦] Claude抄襲中國AI

[Sidney0503 (Sidney0503)]

簡單說 claude 回答說自己是千問  這個迴力鏢就跟我們偉大的黨一樣總是打到自己


https://www.tipranks.com/news/anthropic-faces-distillation-hypocrisy-backlash-as-claude-claims-to-be-alibabas-qwen-ai

短連結 https://shorturl.at/3KcO9

Anthropic Faces Distillation Hypocrisy Backlash as Claude Claims to Be Alibaba
’s Qwen AI

Story Highlights
Anthropic has been accused by developers of distilling Alibaba’s Qwen AI
model.
The controversial claims come after developers found a Claude AI model
identifying itself as Qwen.
Anthropic Faces Distillation Hypocrisy Backlash as Claude Claims to Be Alibaba
’s Qwen AI
Anthropic is facing severe backlash from developers who claim the U.S.-based
AI firm trained its models on Chinese AI systems. Developers say that Claude
Opus 4.8 identified itself as Qwen, the open-source AI model designed by
Chinese tech giant Alibaba BABA -1.97% ▼ , during language tests.



Claim 55% Off TipRanks

Unlock powerful investing tools with TipRanks Premium to make smarter, more
confident investment decisions
Subscribe to TipRanks Smart Investor Newsletter, and discover new investing
opportunities with data-backed stock picks

Ad ends in 14




Video Unpaused

The reports have sparked fresh debate on X, just weeks after Anthropic
accused Alibaba of carrying out its biggest distillation attack against
Claude so far.


Claude’s Chinese Answers Spark Anthropic Distillation Accusations

Claude Opus 4.8 is facing criticism after developers found it named itself “
Qwen” when questioned in Chinese. Screenshots of the exchange spread fast
across X, Reddit and the Chinese developer forum V2EX.


“Looks like @AnthropicAI may have done a little distillation for Opus 4.8…
from Qwen, no less,” Mark Kretschmann, an AI and software engineer, said.


Another developer also said Claude’s Qwen response came up after just three
prompts, with no special settings involved. Commenters also piled on quickly,
calling the timing “embarrassing” given Anthropic’s own history of
accusing Chinese labs of copying Claude.


Other posts argued the episode proves “every accusation is a confession.”
Meanwhile, some Claude testers noted that the Qwen response appeared mainly
in raw API tests, where system prompts are less strict than on the web
version.


Adding to the controversy, another post claimed that Claude was introducing
itself not only as Qwen but also as DeepSeek, a leading Chinese AI company.
The latest findings appear to have disappointed many users even as some
question the validity of the accusations.


So far, neither Anthropic nor Alibaba has commented on the recent reports.


Anthropic’s Distillation Claims Against Alibaba Resurface

The latest controversy follows a letter Anthropic sent the U.S. Senate
Banking Committee on June 10, 2026. In it, the company accused operators tied
to Alibaba’s Gwen Lab of running its largest distillation attack on Claude
yet.


Anthropic also claimed that roughly 25,000 fraudulent accounts produced more
than 28.8 million exchanges with Claude between April 22 and June 5. The AI
firm further alleged that the Chinese government may have been involved in
these attacks, a claim Alibaba has promptly denied.


The letter further said Alibaba’s alleged distillation campaign targeted
Claude’s coding, reasoning and cybersecurity skills. These are the same
capabilities built into Anthropic’s most advanced AI model, Mythos.



However, as similar distillation claims are now being made against Anthropic,
many developers are pointing out the irony of the Claude creator’s own
playbook catching up with it.

--
※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 220.130.178.190 (臺灣)
※ 文章網址: https://www.ptt.cc/bbs/Gossiping/M.1784695560.A.95F.html
推 seabox: 你不能翻譯好再貼嗎
推 thelittleone: 就說中國AI領先米國
→ qber: 魔界抄天堂
推 nsman: 跟中國共產黨一樣
→ pyrolith: 八又翻
→ su4vu6: 中文回答限定
推 DarkKnight: 八又煩
推 z842657913: 就好比跑步游泳一樣 有什麼好抗議的
噓 paericsson: 你不會丟AI翻譯啊
推 andy199113: 哈哈 糗了
→ mit2502: 你問他8*8等於多少  就知道是不是了
推 sr20detll: 文章內容未滿30字(正體中文)
→ gankgf: 八又翻
推 madaofreak: 克勞德被中國AI奪舍啦
→ moccabranco: 就用到別人跑過的資料來訓練啊
推 lipstick: 他們偷了我們的模型
噓 ceab: 連結都不会做？
補上短連結了
推 cage820518: 我怎麼問不出來
推 doig: 跟民進黨一樣，套皮大陸AI說自己的
→ doig: https://i.meee.com.tw/vgVGpgh.jpg
推 neilisme: 美國哪個AI沒有參考deepseek的
→ limitlesscit: 靠北，互抄了喔
※ 編輯: Sidney0503 (220.130.178.190 臺灣), 07/22/2026 15:39:25
→ yesyesyesyes: 不要亂說,這個就是訓練資料在互用
推 al311166: 近親繁殖 不好喔 很容易產生異型
噓 quartics: 那是第三方模型供應商同時有claude以及
→ quartics: qwen 在模型切換中產生的問題
→ sobiNOva: 這時候青鳥知道原因了喔ccc
噓 SkyIsMyLimit: 哪來的野雞文章 想驗證真假你自己問
→ SkyIsMyLimit: 64不就知道了
噓 rodion: 牆臭
→ suckkorea: 全部英文 貼個屁