Ernie Bot 3.5: The new AI champion?

The development of generative AI models is a hotly contested field, and now a new competitor seems to have appeared on the scene. Chinese company Baidu recently announced that their new model, Ernie Bot 3.5, outperformed the well-known chatbot ChatGPT in several tests. These tests are considered important benchmarks in the industry.

Technical details and test results

Ernie Bot 3.5, similar to ChatGPT, is a Large Language Model (LLM) trained with huge amounts of data. According to Baidu, Ernie Bot 3.5 has outperformed ChatGPT in several key benchmarks. In addition, Ernie Bot 3.5 also reportedly beat OpenAI’s latest model, GPT4, in several tests, particularly Chinese language tests.

In tests that simulate standard entrance and qualifying exams, such as those required for admission to a university or to become a lawyer, Ernie scored better in Chinese. However, on a similar test in English, conducted by Microsoft, Ernie fell short of GPT4.

Improvements and future developments

Since the release of the beta version of Ernie Bot three months ago, the program has undergone significant improvements. These include advances in creative writing, plugin access, and code generation. These improvements are visible in creative writing capability, question and answer capability, reasoning, and code generation, as well as in training performance and inference performance.

Outlook

This competition between AI models shows how quickly the field of generative AI is evolving. It remains to be seen how models will evolve and what new capabilities they will acquire in the future. One thing is certain, however: The competition will remain intense, and we can look forward to seeing what innovations come our way next.