Written on the day Wen Xin Yi Yan was released

I was on a business trip recently because of the release of two very important AI products, so it was not convenient for me to write articles. Yesterday, I used my lunch break to write a short article about GPT4, but the Tencent subscription account assistant I used encountered a bug, so the article failed to be saved and was lost. By the way, I want to complain about another Tencent app, Miaojian, which has many bugs and seriously affects the efficiency of writing.

Lack of sincerity towards users and followers

I don't understand why Director Li, a veteran of the Internet industry, is so unconfident. It is understandable to exaggerate in product releases, but have the publishers ever thought that in the past few months, the knowledge accumulation of users and followers in the field of AI has made considerable progress. Are your little thoughts insulting everyone's cognition? As a relevant practitioner, there are a few points that make me very confused. First of all, the model released this time may not be a multimodal model, but most likely just an LLM (large language model). The reason why multimodal is mentioned is because of the sudden backstab of GPT 4. The term "multimodal model" was added later. For example: The essence of Sichuan dialect playback should be the addition of the TTS model to the text, rather than the multimodal itself. TTS is a very mature technology 6 years ago. The various tones in the Amap that everyone uses every day are synthesized by TTS (speech synthesis, text to speech). As for the AI ​​video editing behind it, it seems to be a language summary model plus a CLIP model. This type of application is very common in many video editing software. If you are interested, you can download and install it and try the function of generating videos using pictures and texts. Secondly, this is a semi-finished product. Yes, it is a semi-finished product. The multi-language support that everyone is looking forward to does not seem to have been achieved, and programming language support has not been mentioned. Although Baidu has never been absent from every opportunity to speak out about AI, and the voice is very loud, this time it is extremely disappointing and is still far from everyone's expectations. Thirdly, individual user testing is not open for the time being. I think this is a magical operation, which is simply adding a huge multiplier in front of the two negative terms in front. Don't you even want to give individual users a few minutes of GPU time? If you are a little confident, should you let individual users test it? Things may turn around? Even if Google lays off employees, it still gives AI developers free use of Colab. How did Baidu become so stingy? Finally, there is a video account saying that Director Li revealed that the training of the Wenxin large model was completed by Kunlun Core. I didn't hear the original text, but I think this is the most exaggerated. No matter from which aspect, I think this is impossible to happen, and the reasons come from many aspects. The development of domestic AI chips is worthy of praise, but blindly doing so is boring.

Where have all Baidu’s product managers gone?

I have several very good friends who are product managers at Baidu. I don't mean to criticize them. On the contrary, they are also very good Internet and cloud computing product managers. But in the dimension of products, can Director Li let students who understand products play to their strengths? Let them make decisions? Throughout the release process, we can see that Baidu's new products are struggling to move forward due to capital and commercialization. There are also several issues to discuss here: First of all, in terms of user scenario selection, the demonstration in the video is corporate copywriting generation, but based on the currently open functions, this may still need to be combined with Baidu's existing advertising business monetization, which is too anxious. Therefore, it is not surprising that the final application is open to small and medium-sized enterprise users, not including individual users. Corporate copywriting generation, this does not seem to be a frequently used function. Why don't corporate users who use it frequently choose GPT4? Secondly, with so many individual users paying attention, why not open individual users to trial? Are you really afraid that the GPU cluster will be overwhelmed? If so, wouldn't it be a slap in the face of investors who sold Baidu stocks today with actual capabilities? Maybe the stock will rise tomorrow, but it's unknown? Finally, why do we need to promote the PaddlePaddle framework at the launch event? If your product really penetrates one or several user scenarios, such as having a large model that is sought after by AI developers, will they also like your PaddlePaddle? Is it necessary to promote it so vigorously?

These products can be summed up in one word: patchwork. I also hope that Baidu's product managers will be brave enough to say no to their bosses, can we make some sincere products?

Where has the confidence of Baidu’s engineers gone?

The former Baidu engineers were like gods in the Internet. I have many such technical masters around me, and they are all very good friends. But today, these successors? Where is your confidence? Your lack of confidence was conveyed to everyone in front of the screen by Director Li through the live broadcast room. Where is the arrogant Director Li a few years ago?

Why are you being led by Open AI?

In addition to Baidu, there are many other teams in China that have released their own multimodal large models, including: DAMO Academy, Zhiyuan Research Institute, Huawei, etc. Moreover, these large models have been released for at least more than two years, and the training of large models should be familiar to them. However, when OpenAI's ChatGPT became popular, why did everyone turn to LLM (large language model)? Are they not confident? However, what is more ironic is that when GPT4 announced multimodality, we turned around and pursued multimodality again. May I ask, have we thought clearly about what kind of product we want to make? Or is it just for publishing papers or Pr?

When we admire Altman, have we ever thought about what kind of abilities he has?

First of all, it must be clear that Altman is a product manager with programming skills, a business genius, a savvy investor, etc. However, he has been portrayed as a genius programmer by domestic self-media. Wake up, everyone. Behind the popularity of OpenAI, it is Altman's business and product design. Take the time to listen to Altman's analysis of business. Knowing his talent background, you must be sure that both ChatGPT and GPT4 are well-thought-out commercial products, not a pile of technology.

Baiduers, do you no longer believe in the Internet?

Before the release of Wenxin Yiyan, Baidu had already started promoting Wenxin Yiyan's enterprise service through business marketing. Today's release is really jaw-dropping. An Internet product trial requires filling out such a complicated registration form and waiting for approval? Have you forgotten how easy Baidu was to use 20 years ago? Does it need marketing sales? Do users want to come to Beijing with money to find you? But today, have you forgotten the power of the Internet? It can push cloud services to users at the speed of light. Why don't you promote your products through word of mouth? Is the cost of burning GPUs higher and less efficient than the cost of business marketing? How did Altman do it? Which Internet product became popular through marketing?

As a winner of Toutiao's Qingyun Plan and Baijiahao's Bai+ Plan, the 2019 Baidu Digital Author of the Year, the Baijiahao's Most Popular Author in the Technology Field, the 2019 Sogou Technology and Culture Author, and the 2021 Baijiahao Quarterly Influential Creator, he has won many awards, including the 2013 Sohu Best Industry Media Person, the 2015 China New Media Entrepreneurship Competition Beijing Third Place, the 2015 Guangmang Experience Award, the 2015 China New Media Entrepreneurship Competition Finals Third Place, and the 2018 Baidu Dynamic Annual Powerful Celebrity.