If you like this work, please click "Follow" at the top right. Thank you for your encouragement and support, I hope to bring you a comfortable reading experience. This article, only to do the first creation of today's headlines, not released on any other platform, the original is not easy, plagiarism, manuscript washing will be deeply studied.
Digital Review: Comprehensive Evaluation of AI Language Large Model
Over the past year, OpenAI has launched ChatGPT, a language model that exploded in the tech world. Not only can it chat with people, but it can also do arithmetic, programming, and even help programmers fix bugs.
However, not to be outdone, tech giants have invested in competitive development to ensure they are not eliminated.
Microsoft is one example. Shortly after the release of ChatGPT, Microsoft quickly launched NewBing, applying GPT-3.5, a powerful model of human language understanding, to its own search engine. This makes the original general Bing search engine become an almost omniscient and all-powerful AI search engine, with functions such as online search and contextual communication.
Relatively speaking, the AI chatbot Bard released by Google performed much less.
Not only were incorrect answers at the conference, but even users of the closed beta version asked when they closed down, which was also incorrectly quoted. This hints that Google may not be training enough. Despite this, due to external pressure, Google had to launch this product with limited training time.
It's worth noting that Google didn't officially support Chinese language until July 13 this year. However, Baidu's Wenxin Yiyan and iFLYTEK's Spark model have been opened for testing one after another, and some users have even begun to compare them sideways.
In order to evaluate the capabilities of these large language models, we will comprehensively evaluate the real-time information search, Chinese comprehension ability, and multimodal recognition ability.
Real-time information search capabilities
For digital issues, such as the release time and hardware configuration of the Honor Magic V2, there are differences in the performance of each model. Bard is somewhat abstract in his answers, and accuracy is problematic, possibly due to a lack of understanding of Chinese. Wen Xin's performance is relatively stable, with less misinformation, while the Spark model performs the worst in this regard, and even some wrong information appears.
Chinese comprehension
On the Chinese Comprehension Test, Bard performed relatively poorly, with insufficient understanding of some Chinese slang and logical relationships.
Wen Xin Yiyan and the Spark model perform better in Chinese understanding, especially the Spark model can also provide practical examples to help users understand.
Multimodal recognition capability
Bard has strong multimodal recognition capabilities, especially in image recognition and programming. However, its Chinese ability is weaker compared to other models. The Spark model performs poorly in multimodal recognition and does not provide image upload function.
Comprehensive evaluation
After a major update in July, Google's Bard has been able to be used relatively stably in Chinese environments. It supports Chinese chat, image recognition feedback, and the ability to search for real-time information online.
Although it is rich in functions, on the whole, the performance of each language model is not much different, and they have not broken through the framework of "chat question and answer".
Finally, we still expect these language models to create killer use cases in more fields to achieve greater breakthroughs in technology. Whether it is Google's Bard, Baidu's Wenxin Yiyan and iFLYTEK's Spark model, it is likely to bring more surprising innovations in the near future.
We sincerely invite you to click the "Follow" button to continue to push such articles for you in the future, thank you very much for reading and supporting, and hope to interact and communicate with you more.