HONG KONG (Reuters) – China’s leading search engine provider Baidu Inc said the latest version of its ChatGPT-style service outperformed the widespread Microsoft-backed OpenAI chatbot on several key metrics. Announced.
Baidu said in a statement on Tuesday that the latest version of its Ernie AI model, Ernie 3.5, surpassed “ChatGPT on overall proficiency scores” and surpassed “GPT-4 on several Chinese proficiencies.”
The Beijing-based company was conducted by state-owned newspaper China Science Daily using a dataset containing AGIEval and C-Eval, two benchmarks used to evaluate the performance of artificial intelligence (AI) models. I quoted the test I did.
OpenAI did not immediately respond to an email seeking comment outside of normal business hours.
Baidu’s announcement comes at a time when the global AI buzz that started with ChatGPT has spread to China, with domestic companies launching competing products one after another.
Baidu launched the language AI Ernie Bot in March, becoming the first major Chinese tech company to debut an AI product to rival ChatGPT. Built on Baidu’s old Ernie 3.0 AI model, Ernie Bot has been tested on an invite-only basis for the past three months.
Other Chinese tech giants, including Alibaba Group and Tencent Holdings, have since revealed their AI models.
Baidu says the new model has improved training and inference efficiency, allowing for faster and cheaper iterations in the future. Baidu also said the new model will support external “plugins”.
Plugins are additional applications that allow Baidu’s AI to work in more specific scenarios, such as summarizing long texts or generating more accurate answers.
ChatGPT started supporting plugins in March.
(Reporting by Josh Ye, Editing by Himani Sarkar)
Copyright 2023 Thomson Reuters.
