With the advent of domestic top-level ChatGPT large models such as Wenxin Yiyan and Tongyi Qianwen, artificial intelligence with language understanding and generation capabilities is leading the innovation and development of the industry. As a leader among artificial intelligence companies, HKUST Xunfei has also begun to join this competition.
On April 20, HKUST Xunfei announced that it will officially release its latest "Spark" cognitive model on May 6, and disclosed the latest progress of the model in detail at the performance briefing that night. This move has aroused great concern from the whole industry. It is reported that "Spark" is developed around a series of practical scenarios such as "knowledge question and answer, code programming, mathematical calculation, creative association, language translation". User needs understanding and task execution.
And just a few days later, HKUST Xunfei’s “Spark” cognitive model opened the internal test channel, and I was lucky to get the internal test quota. Next, I will actually test the Chinese version of ChatGPT “Spark”, what kind of experience will it have? Woolen cloth?
First of all, how to experience the internal test large model?
1. Search "HKUST Xunfei" through Duoyu browser, and directly enter the experience page, which is more convenient for me.
2. You can enter the website (https://xinghuo.xfyun.cn/) to enter the official website.
Olympiad test:
I specially found a lot of Olympiad questions to test, including the most popular problem of chicken and rabbit in the same cage. Its response is extremely fast, the problem-solving ideas are very clear, and the entire calculation process is fully displayed. Compared with many question-searching software, such a fast response is really much faster.
Writing Proficiency Test:
"Writing ability" is a basic ability that a large model must have. Since the recent BMW mini ice cream incident was condemned by netizens on the Internet, I asked it relevant questions about this incident. It answered very standardly and could fully understand From the beginning to the end of this matter, it can further help me refine the problem. The text writing ability is strong, and the whole process is complete.
However, the completion of a certain work still needs to be improved. For example, I asked it to write a seven-character poem expressing the deliciousness of milk tea. Although he could write it according to the standard format, the rhythm in it could not be called Above is a real poet.
Daily Questions Test:
I asked some questions in life, such as "how to make fish-flavored shredded pork", "can dish soap be used to wash fruit", the answer given by Xinghuo not only listed detailed steps and methods, but also " A variety of reminders are attached in a thoughtful manner, and the structure is very complete.
In addition, I also specifically asked about "whether men/women need to do housework", and his answer was very pertinent. On gender issues, his position respects the equality of men and women, and does not create the logic of gender antagonism.
Code problem test:
Like ChatGPT, "Xinghuo" can also generate codes, such as the "Zibo Cuisine" crawler program generated as shown in the figure below. As for whether it can work, professional technicians have to judge it hahaha
Logical Questions Test:
I have used several domestic AIs, and they are relatively easy to overturn in terms of reasoning logic, but to my surprise, the performance of the Spark large model is still good, and it is not bypassed by the problem.
Internet hot-stalk test:
In the end, I decided to ask the fresher Internet memes about "Xinghuo" to see if they could answer.
I found that in the face of some "stalkers" on the Internet, "Xinghuo" can't be well understood. But when it is officially released in May, it should be iterated to fix this kind of problem.
Through a series of evaluations, I found that the overall performance of the "Spark" model is very good. Especially in daily application and learning work, "Xinghuo" has a very stable performance, showing excellent performance and potential, and injecting fresh blood into the field of domestic large-scale models.
It is worth looking forward to that "Spark" will be officially released on May 6, which will undoubtedly bring more surprising experience to many daily users. As a "new army" in the industry, the "Spark" model has great potential and prospects, and will help promote the application and development of artificial intelligence technology. I believe that in the future, "Xinghuo" will become an indispensable and reliable intelligent assistant and partner for users.