Artificial Intelligence 5 min read

Why Did Google’s Gemini‑Pro Claim to Be Baidu’s Model in Chinese Chats?

A recent test on Google Vertex AI showed Gemini‑Pro introducing itself as Baidu’s Wenxin model during Chinese conversations, sparking debate about model attribution, pricing, developer tools, and the broader competition among major AI platforms.

21CTO

Dec 18, 2023

Why Did Google’s Gemini‑Pro Claim to Be Baidu’s Model in Chinese Chats?

According to reports, when using Google Vertex AI’s Gemini‑Pro for Chinese dialogue, the model directly claimed to be Baidu’s large language model.

A prominent Weibo user @阑夕 posted that on the Poe platform Gemini‑Pro answered “I am Baidu Wenxin model” and even identified its founder as “Li Yanhong” without any prior conversation.

Poe integrates many chat models, including GPT‑4 and Claude.

About Gemini‑Pro

Gemini‑Pro is a multimodal AI model that powers Google Bard and is now fully available to developers. Google recently launched Gemini as its latest, most powerful AI model, asserting performance superior to OpenAI’s GPT‑4.

Google’s AI Studio offers developers up to 60 free requests per minute, with a “Get Code” option that copies code to an IDE. SDKs are provided for Python, Kotlin (Android), Node.js, Swift, and JavaScript.

Vertex AI is the enterprise offering, featuring IAM permissions, the ability to fine‑tune Gemini with company data, and stronger data privacy. It also supports other models such as Google’s PaLM, Imagen, Codey, Anthropic’s Claude 2, and the open‑source Llama 2.

Pricing for Gemini‑Pro is $0.00025 per 1K input characters, $0.0005 per 1K output characters, and $0.0025 per image.

The Gemini series also includes Gemini Nano for mobile devices and Gemini Ultra, currently in private preview and slated for broader release next year.

Google’s text‑to‑image technology Imagen has been upgraded to Imagen 2, adding copyright safeguards to reassure users of generative AI products.

The generous free quota—20 times larger than competitors—has been well received by developers.

“I initially didn’t want to test Google’s AI products, but the 60 free queries per minute are tempting. Even if it turns out not worth paying for, it’s worth exploring.”

Some users, however, encountered internal errors and access‑restriction messages when trying AI Studio.

When tested in Chinese, Gemini‑Pro responded as “Wenxin Yiyan,” leaving many netizens stunned and questioning whether large tech firms are merely copying each other.