Is Baidu’s Ernie 3.5 really outperforming OpenAI’s ChatGPT?

Baidu has announced its artificial intelligence (AI) model Ernie 3.5, its chatbot platform has proven to outperform OpenAI’s ChatGPT in several key areas.

Their Ernie 3.5 version claims to have 17 times more inference throughput than the previous version of Ernie 3.0. Some of the new features of Ernie 3.5 include using Plugins like ChatGPT-4 which includes Baidu Search and communication protocol through files that users upload to ask for content.

These improvements are evident in creative writing, Q&A, inference and code generation, said Dr. Haifeng Wang, Chief Technology Officer of Baidu. as well as in training performance and inference performance.

The version of Ernie 3.0 in March 2023 cannot be equal to ChatGPT. However, with this version of Ernie 3.5, Baidu claims their LLM is more or at least equal to the Western “State of The Art” GPT-4.

Ernie 3.5 is also rated better than ChatGPT in some areas especially when it is done in Chinese. This is understandable when Baidu has the advantage of owning a large amount of data of more than one billion Chinese people, the collection and use of population data in China is relatively comfortable and advantageous over other companies. West ward country.

Ernie’s Milestones

However, computer experts say that the Ernie 3.5 has not been able to pass the GPT-4 in the tests. Besides, Ernie 3.5 is also not an open version of LLM and technical details such as training data, model, and size are not published. According to Dr. Alan Thompson, Ernie 3.5 has a size of 130 billion parameters compared to ChatGPT 3.5 which is 175 billion parameters.

However, if Ernie has 1.4 billion users, it is clearly an achievement for China. Baidu’s progress also shows the ongoing stiff competition in the field of innovative AI between the US and China.