Intro#
As various internet companies of all sizes have entered the AIGC track, AI is advancing rapidly, and other technologies are also experiencing a major breakthrough. The cost of learning time and the requirement for knowledge reserves are high, making it difficult for me to update frequently. Therefore, I am creating this new column to periodically share the latest AI and other technology. The main focus will be on sharing interesting projects and technology news that I find. This is the first issue.
Xiaomi AI Large Model MiLM-6B#
According to GitHub, Xiaomi, which announced its entry into the large model track a few months ago, has developed "MiLM-6B" with a parameter model of 6.4 billion, achieving the best results in both C-Eval and CMMLU among models of the same size.
By today's standards, a model with 6.4 billion parameters is still relatively small. It is uncertain whether this top ranking is due to their excellent algorithm or the small number of parameters leading to errors. Another possibility is that Xiaomi has its own way of operating and has special optimizations for these two propositions. The specific level still needs further training, and it will only be known after testing when the parameter volume becomes larger and is made available.
Although I have some doubts about this, I am also very optimistic about Xiaomi's AI project because it is obvious that Xiaomi's AI is aimed at upgrading "Xiao Ai". There is no need to become an all-around warrior like "ChatGPT". The focus should be on the interaction between humans and intelligent devices. After all, "Xiao Ai" is already in the top tier of mobile AI. So I am still very optimistic.
Zhidu Technology to Release Dual AI Smart Speaker ChatMini#
According to the official WeChat account of Zhidu Technology, the world's first smart speaker with built-in ChatGPT, Vifa ChatMini, will be launched globally on August 17th. The official website shows that ChatMini will also be equipped with Baidu's AI large model "Wenxin Yiyu". With the support of dual AI, the official price on the website is set at $259.
I personally want to support domestic projects, but when I see the keywords "ChatGPT" and "domestic manufacturers", I can't help but feel a chill. We all know that even ChatGPT mirrors are difficult to survive in China now, not to mention the basic parameters of a speaker. How can a speaker with ChatGPT be used normally? How can it survive? Will the overseas version use ChatGPT while the domestic version uses Wenxin Yiyu? It's terrifying to think about it.
Let's wait for the press conference on the 17th for the specific details. It will be held at 20:00 in the evening. The smart speaker will also make its debut on Douyin's Lv Xiaotong, Zhidu Technology & Guoguang Electric Chairman Lu Hongda, Denmark's Vifa speakers, and other live broadcasts, as well as Vipshop's Vifa live broadcast room. Interested friends can watch it.
Baidu Search Engine Integrated AI#
Users who have joined the internal testing of Baidu's Wenxin Yiyu can now find a conversation feature in the top left corner when using Baidu's search engine, similar to Newbing. Because I rarely use Baidu myself, I just discovered this while writing this article.
After a brief look, it has all the necessary functions, and the application of the Inspiration Center is very rich. It is similar to assigning roles to ChatGPT, except that this role has already been defined for you. Since it is still based on Wenxin Yiyu, I don't have high expectations for the level of conversation and application. I won't test it myself, but interested friends can test it themselves. I guess there won't be much difference between it and Wenxin Yiyu.