In the field of artificial intelligence, the "hallucination problem" of large language models has been a key challenge troubling the industry. Recently, a test based on the HHEM-2.1-Open evaluation system showed that the GLM-4-9B model from Zhihui AI achieved remarkable results.
Test data indicates that among the 85 large language models evaluated, GLM-4-9B topped the list with a 98.7% factual consistency rate and a 100% response rate. Notably, the model's hallucination rate is only 1.3%, which not only ranks first but also surpasses the GPT series from industry giant OpenAI and Google's Gemini series models.
Image Source Note: Image generated by AI, image authorized by service provider Midjourney
This groundbreaking achievement marks significant progress in reducing the "hallucination rate" of China's large language models, bringing new possibilities for enhancing the reliability and accuracy of AI applications. It not only reflects Zhihui AI's technical strength in model development but also sets a new benchmark in the global AI field.
In the context of increasingly fierce competition in AI technology, the outstanding performance of the GLM-4-9B model will undoubtedly draw more attention to Zhihui AI, while also granting Chinese AI companies greater influence on the international stage.
