Author: Zhao Tianyi Editor: Lin Jiongjia "As long as you can speak, you can become a developer!" Robin Li confidently announced at the 5th Baidu AI Developer Conference. In the past year, there have been many stories of AI breaking the circle, and urgent changes such as "AI has made writing a zero threshold" and "you can be a painter without drawing" have been constantly appearing in our ears. And this time, it is programming that becomes "zero threshold". On April 16, at Baidu's annual AI developer conference, Robin Li focused on three "out-of-the-box" AI development tools - AgentBuilder (agent development tool), AppBuilder (AI native application development tool) and ModelBuilder (model customization tool). According to the official introduction, AgentBuilder is open to everyone and can be easily built without programming foundation. AppBuilder and ModelBuilder, on the other hand, are geared towards those who have a certain coding foundation. In the future, these two tools will also open up the link and jointly serve industrial-level enterprises. Robin Li, who was on stage, spent half of his speech introducing these three tools, which shows that he has great expectations. However, PConline's interview with dozens of practitioners and college students in the AI town that day was a different story: developers didn't seem to buy it.
"I don't feel like I was pleasantly surprised by this [developer conference]. At present, these can be said to be the integration of existing functions, and there is no breakthrough from 0 to 1. Mr. Wu, who has been paying attention to the development of the AI industry, said to PConline. Mr. Wu is not the only one who has such a "thirst-unquenching" feeling. "It's a shame that a lot of the content today is screen recorded. After listening to Robin's speech, I also went to the exhibition to listen to the detailed introduction of the staff, and found that it was still a large model of dialogue. A student from Tsinghua Research Institute shared his feelings with PConline.
Robin Lee presented on stage the Singapore Tourism Board as the first case of AgentBuilder. In the exhibition, when faced with the question of "when will the least number of people go to Singapore", the answer given by the agent of Singapore tourism is "the least number of people in January ~ March". When PConline searched for the question, Baidu's page gave the answer "April, May, and November".
On the other hand, an AI industry insider shared that there are already many AI businesses like AgentBuilder in the market, and Baidu's gameplay is not new. "This kind of AI business can be specially used by companies in e-commerce, education, consulting and other industries, and has a very significant effect on improving the conversion rate of leads and reducing the conversion cost of effective leads. At present, most of the market is one-to-one private customization with merchants. Such an agent is equivalent to a "high-level version of Taobao customer service", which can have a dialogue with users and provide some simple functions and services. According to AI industry insiders, according to the current situation in the market, the technical team with single-digit personnel will use the open-source model and fine-tune it after receiving the demand, and the development of a custom agent can be completed in more than a month. Merchants do not need any operation, they can directly install such an agent "customer service" in WeChat, on the webpage, and even on various social platforms. The service fee is calculated based on the customization fee plus the "footfall" of the agent. On the day of the developer conference, Baidu's stock fell 2.69% and is still in a downward trend as of press time. Notably, Baidu's share price has fallen nearly 25% from its highs so far this year. The repeated cold dilemma seems to have become a nightmare that Baidu cannot escape.
Since the release of ChatGPT, Baidu, as one of the earliest Internet companies to deploy AI, has been burdened with heavier and heavier expectations. Four months after the release of ChatGPT, Wenxin Yiyan kicked off the prelude to the domestic large-scale model scuffle. But at that time, Wenxin's words were still difficult to be convincing, and the pictures of "confident" and "husband and wife lungs" also aroused ridicule on the whole network. The shadow of the "wrong version" began to loom over Baidu. At this developer conference, Robin Li particularly emphasized that "the number of users of Wenxin Yiyan has reached 200 million in one year". In one year, Wenxin has completed the iteration from 3.0 to 3.5 and now 4.0. On the AI industry's largest "table", Baidu must not be left behind.
The industry environment does not allow Baidu to slow down. Alibaba's Tongyi Qianwen, Tencent's Mixed Yuan Assistant, and Byte's Doubao...... Everyone is striving to be the first, using AI to reinvent themselves and expand their frontiers from the inside out. Just as there was fierce competition at the time of the rise of the mobile Internet, today's AI is also a zero-sum game. However, Baidu's opponents are not only "Internet veterans". Nowhere is this more evident than in the AI world, where startups have a better chance of being on top of the wave. The success of OpenAI, Midjourney in the AI painting world, etc., have all defeated trillion-dollar tech giants with a very small number of people. The "new life" in China is also as fierce as a tiger. According to the data of the AI product list, in March 2024, Baidu Wenxin Yiyan will have 1490w visits, while the second place is kimichat, which was developed by the dark side of the moon, with 1261w. The gap between the two is extremely small, and Kimi's access data is still on the rise. In early March, Kimi's founding company, Dark Side of the Moon, raised more than $1 billion in funding. This start-up has detonated a new wave of enthusiasm in the capital market, and even the "Kimi concept stock" has risen with the tide. Baidu's first-mover advantage doesn't seem to be that important anymore. At present, the number of large language models that have been released in China has exceeded 300. Globally, the competition is only going to get more brutal. In March 2024, ChatGPT received 1.86 billion visits, more than 120 times that of Wenxin Yiyan. "On the one hand, ordinary users have too high expectations for AI, thinking that AI should be completely intelligent like in the movie, and the key technology has not yet reached that level. An AI entrepreneur said to PConline, "On the other hand, everyone can also see what level foreign countries are, and they are naturally unwilling to pay for it when compared with their hands." "The gap between high expectations and the reality of technology has always been two mountains on Baidu's shoulders.
Robin Li mentioned in his speech that Baidu "stepped on countless pitfalls and paid high tuition", and according to the practice summary of the past year, the development path of AI native applications was positioned as MoE structure, small model fine-tuning, and agent mechanism. The MoE structure refers to the new trend in the market - the Mixture of Experts. In the process of AI landing, the industry has gradually found that "large and complete" large models are not suitable for landing. Generally, the requirements are small, and when the large model is invoked, too many resources are consumed, and the cost performance is too low. Instead of relying on one model to solve all problems, hybrid calls are used to determine when to call a "small but beautiful" small model and when to call a large model based on the problem, which saves resources, reduces costs, and improves response speed. This development idea has been recognized by many AI startup teams. Zhou Ze'an, the founder of ChatPPT, said in an interview with PConline that the foundation of their products is to use the MoE structure, and the self-developed small model cooperates with the Wenxin Yiyan large model to jointly provide users with high-quality services. The direction is right, but the development path of AI native applications is not easy, and its dilemma is more in promotion and operation. In this regard, Zhou Ze'an has a deep experience: "In fact, the main (problem) of domestic AI native applications is traffic. Either you have a lot of capital and you can buy traffic. But if it is built from scratch by an entrepreneurial type, almost no one will use it without traffic, which is why there is no AI application that has broken the circle so far. "At this point, Baidu is ahead of the curve. Baidu's integrated Wenxin intelligent twin platform is designed to provide developers with a channel to monetize traffic. In ecological products such as Baidu search, maps, and Tieba, there will be the exposure of agents, and we will try our best to solve the traffic worries for developers. "What Baidu is doing now is building systematically, which is very necessary. It's normal to feel like you're drawing a pie, because the whole system needs to be done by us developers. A programmer who is trying to develop independently shared. While talking to PConline, he was in front of a laptop and was excited to try out AppBuilder. Baidu, as the number one person of the old "BAT", has undoubted strength and resources. But if it wants to return to the core position of the technology industry through AI, Baidu's pace must be more stable and faster.
END
AI anti-deception· PC big trick 🚨AI! I was deceived by accident?!
🔥AI scams are getting rampant, and you're still unaware?
🤔 How good are your scam prevention skills?
👇 Poke at the bottom to see if you can see through the AI's cunning trap?
👇 Identify the QR code below to get more AI anti-scam tips!