Author | Xue Xingxing

Editor | Jiang Jiao

Cover | "Silicon Valley"

On July 7, Baidu held an internal director's meeting. This communication meeting, known internally as the "Fireside Chat," is held quarterly and attended by Li Yanhong and Baidu's internal management personnel at the director level and above. It is not open to grassroots employees. On July 12, Baidu internally released the minutes of the director's meeting.

A document exclusively obtained by "The Hill" reveals that in addition to Li Yanhong's oft-repeated AI strategic issues such as the superiority of closed-source large models and the direction of AI applications with intelligent agents, he also rarely responded to Baidu's current governance challenges at this meeting, including low morale within the company, non-accountability of senior executives, and difficulties in cross-departmental collaboration.

Two months ago, Qu Jing, Baidu's head of public relations, resigned due to several controversial short videos, sparking questions about Baidu's culture both inside and outside the company. This Fireside Chat was held for the first time after Qu Jing's resignation. The meeting was a dialogue between Li Yanhong and Liang Zhixiang, Qu Jing's former supervisor and currently the senior vice president in charge of Baidu's public relations business.

Li Yanhong did not mention Qu Jing at the meeting. In response to a question about what constitutes an excellent middle and senior manager, he said that managers must have leadership and set an example for their subordinates. "Culture is not something that can be written down or articulated; it's how you solve problems when they arise," he said.

He said that managers who do not understand the industry and cannot make decisions should not be in their positions. At Baidu's Advanced Employee Awards ceremony in early May, Li Yanhong said that he is a relatively good-tempered boss who rarely gets angry with employees, but he also has a bottom line that must not be crossed.

In Li Yanhong's latest quarterly OKR, O4 mentions the timely matching of organization and strategy, KR1 mentions learning AI-native thinking to further focus on the company's strategic direction, and KR2 mentions timely summarization of experience and lessons learned to align cross-organizational communication. Li Yanhong's explanation of the OKRs did not involve the Qu Jing incident, saying that they were all set based on AI directions.

01｜"This time will be different"

In the first half of 2023, riding on the wave of large models, Baidu's share price continued to rise, surging to HK$154.4 in March, the highest point since early 2022. However, it has since declined steadily. Even though it recently surged due to Luobo Kuaipao, setting the highest single-day gain in over a year, it has not yet returned to last year's high. Over the past year, Baidu's Hong Kong share price has fallen by 37%.

Some market analysts believe that Baidu's current AI strategy has not yet significantly driven its main business, and online advertising remains the backbone of Baidu's revenue. This business has seen slowing growth for several consecutive quarters, and there are still no signs of recovery in the second quarter. Li Yanhong said on Baidu's first-quarter earnings call that the company has been commercializing generative AI for its advertising business since the second half of last year, currently bringing in millions of dollars in incremental revenue per quarter.

At the internal director's meeting, Li Yanhong also rarely admitted that Baidu's growth rate has declined, saying it is a "difficult time." He hoped that employees would not be swayed by outside opinions and instead focus on delivering results, with other things following naturally. "If you believe in this, stay and let's work together. If you don't, there are better options outside, that's it," he said.

Li Yanhong said that the current state of the mobile internet has become a zero-sum game, and Baidu is in a passive position. "Our content ecosystem still lags far behind that of Douyin and WeChat. It's not easy to catch up using the same mobile internet strategies," he said.

He said that Baidu's morale is indeed low at present, which is due to the company being in a transition process. Past technological breakthroughs led them to mistakenly believe that the era of AI had finally arrived, but "the best outcomes have often been disappointing." He believes that "this time will be different," but it remains to be seen how long it will take to usher in the true era of artificial intelligence.

Li Yanhong feels that the anxiety and urgency felt by employees at present are normal. He cited the example of Lixiang MEGA, saying that their sense of anxiety is even stronger than that of Baidu. Li Xiang addressed his company in the corridors in 2022, saying that they had been targeted by the world's most aggressive companies. However, even so, situations like the unsalable MEGA still arise, so a sense of urgency is necessary.

His solution is "focus." Li Yanhong said that Baidu cannot fight too many battles, as the company's resources are limited. They must allocate resources to areas truly related to core businesses to avoid resource dispersion that could lead to defeat in battles that should have been won.

A former Baidu IT employee told "The Hill" that Baidu used to be a company that was particularly willing to engage in new businesses. "Because it didn't have any particularly outstanding businesses, to put it bluntly, I could start an independent project, and it would take off, but basically, none of them did," he said.

Over the past decade, Baidu has tried many cross-border businesses beyond AI and search, including food delivery, group buying, e-commerce, live streaming, short videos, and more. Before returning to Hong Kong for its listing, it planned to spend $3.6 billion to acquire YY to compensate for shortcomings in its mobile ecosystem. However, this acquisition, which was Baidu's largest ever, was terminated after several years of delay.

Before large models, its most recent abandoned strategic direction was the metaverse. In 2020, Baidu launched "Xirang," China's first metaverse product, and even held its AI Developer Conference the following year within Xirang. However, Xirang performed poorly, with graphics that were inferior to mobile games from several years ago. Last year, Xirang's head, Ma Jie, announced his resignation, which was seen by outsiders as a sign that Baidu was abandoning the metaverse.

Baidu Xirang

"Startup projects came one after another, but none of them were particularly useful. Basically, none of them took off, and they all failed in the end," said the former Baidu IT employee.

Li Yanhong feels that it is normal that many of their battles are not won, as they are doing many things that have not been done before. Without a way to prove whether a path is viable or not, they go to war. "After three to five years, we may find that this path is not viable, which is not uncommon, and I accept that because no one in the world has walked this path before," he said. They are at the forefront and must acknowledge this rule. They must achieve nine failures and one victory.

02｜Managers must have leadership and set an example for their subordinates

Baidu's core values are "simple and reliable," which have been in use for over 20 years, but they still do not seem to have deeply imprinted organizational culture onto every employee like Alibaba has. One Baidu employee told "The Hill" that they have no understanding of these words. Another former employee said that internal employees even have a sarcastic saying about it: "Simple, can rely (on someone else)."

The Qu Jing incident in May further intensified questions about values both inside and outside the company. In early May, Qu Jing, Baidu's head of public relations, posted several controversial videos on short video platforms, attracting public attention. Qu Jing later apologized on her personal social media account and soon resigned.

Cui Shanshan, Baidu's senior vice president, said at the company's Advanced Employee Awards ceremony that it is not surprising for a company nearly 25 years old to encounter problems, but it is wrong not to view them soberly and rationally. She said, "At Baidu, nothing is more worth upholding and defending than the values of simplicity and reliability." Li Yanhong also said at the event that these outstanding employees being honored are the true representatives of Baidu.

Li Yanhong at Baidu's internal awards ceremony / Source: Baidu

Several employees interviewed by "The Hill" mentioned management issues within Baidu, such as middle-level leaders' inaction, buck-passing, and only following orders from above. "What the superiors want, I do," one said.

One of the questions at the director's meeting was about the criteria for excellence among Baidu's middle and senior managers. "There are many high-T, high-P, and even M4, M5 managers in the company who are 'sitting in the temple, unaware of the distant world.' They do not take responsibility at critical decision points, requiring business departments to do a lot of internal science popularization work, leading to low efficiency, long processes, and slow decision-making. Robin also said at Baidu's Pride Awards ceremony that these outstanding employees represent Baidu's cultural values. So, what are the criteria for excellence among middle and senior managers?"

Li Yanhong replied that some of the directors or M4, M5, and even T10, P10 and above-level employees have gradually become detached from specific work. He said he has requirements for HR to assess leaders, and if a leader does not understand the industry, cannot make decisions, or cannot take charge, they should not be in that position. For executors, their requirement is that "whoever benefits should drive it forward, and if it cannot be resolved, escalate it up the chain."

"Both leaders and employees must understand that there will always be a conclusion and a solution to any matter. If no one can resolve it in the end, it will come to me, and I will give you an answer and a decision. Nothing can be left unresolved and dragged on. So, do not let this matter delay our decision-making and execution," Li Yanhong said.

Li Yanhong said that in addition to the requirement of "simple and reliable," he also has requirements for managers' leadership. "Your subordinates look up to you, so you must set an example for them," he said, addressing Baidu's middle and upper-level employees at the director level and above. "In the position of Baidu director, you must be able to set such an example, so that more people feel that this is a positive, competitive, and promising company."

"Baidu used to be a bit like a state-owned enterprise," the former Baidu IT employee told "The Hill." But now, Baidu has started to reduce costs and increase efficiency, with strict attendance management. One Baidu employee said that the company tracks attendance times, with employees expected to arrive at 9 a.m. but usually leaving around 8 or 9 a.m.

A former Baidu employee who was laid off recently said that at first, he did not know whether his layoff was the decision of HR or his department leader. Later, they analyzed privately and believed it was the AI's decision, "calculating your off-work time, your code submissions, and your overall cost-effectiveness."

Layoffs have been ongoing. According to Baidu's annual report, the company had a total of 39,800 employees at the end of 2023, down from 41,300 at the end of 2022, a reduction of 1,500 employees over the year. However, this figure is still smaller than the number of layoffs at Alibaba and Tencent.

The former Baidu IT employee said that in previous years, Baidu's R&D positions at T6 and T7 could directly transition to management and become M-level, which offered higher salaries and was easier to advance, leading to a decline in coding among many employees. However, against the backdrop of cost reduction and efficiency enhancement, "they began to cut these management positions, especially those who do not code." In the past, Baidu employees received an annual salary adjustment of around 6%, which has now been abolished.

03｜If you don't speak up, you'll be misled

The past period has been one of the most intense times for Li Yanhong to speak out externally. He has continuously talked about AI strategies, sparking many controversies, such as "closed-source large models are definitely better than open-source ones." His latest view is that intelligent agents are the promising direction of AI applications, and search is the largest entry point for intelligent agent distribution.

Li Yanhong said that he was the first guest speaker at the industry forum of the Shanghai Artificial Intelligence Conference, and before speaking, he anticipated that many of his statements would be controversial. However, the current AI market is chaotic, "with people from all walks of life expressing various opinions on AI. If we don't speak up, we will be misled by others," he said, so he continues to output externally, trying to influence outside understanding of the current state of AI development and future directions.

Li Yanhong at the Creat 2024 Baidu AI Developer Conference / Source: Baidu

Over the past year, almost all of Baidu's businesses have integrated AI. At last October's Baidu World Conference, Li Yanhong announced that all Baidu applications have been reconstructed through AI, including Baidu Search, Baidu Wenku, Baidu Netdisk, and more.

The former Baidu IT employee said that the current employee promotions and business priorities are all considered in relation to AI. Some business lines that were originally far from AI have had to think about how to integrate AI. "When it comes to promotions, those who are related to AI will advance faster, it's as simple as that," he said.

At this director's meeting, Li Yanhong reiterated his views on AI strategy, including his previously discussed topics such as closed-source being better than open-source, intelligent agents, and large models having a broader prospect in ToB than ToC. He also mentioned multimodality, Apple Intelligence, and price wars among large models.

Multimodality is the right direction, and EB5 will be compatible with multimodality

"I believe that GPT-4o is the right direction. Before that, everyone had been eagerly awaiting the release of GPT-5. But after GPT-4's release, we saw GPT-4V, GPT-4Turbo, GPT-4o, all sorts of attempts based on GPT-4. The 4o direction is multimodal, with faster response speeds, more natural connections, and a more native experience. It can process various images, sounds, texts, etc., within a single model, enabling better user experiences and applicability in more fields. I quite agree with the 4o development direction," he said.

Li Yanhong said that they have also heard some voices regarding multiple business needs. Regardless of whether they will release EB5 or any other new models in the future, they will ultimately be compatible with multimodality.

Apple Intelligence has little to do with Baidu

Apple Intelligence has little to do with Baidu, as it mainly emphasizes large models on the end side, which Baidu has little control over. Domestic brands like Huami OV and foreign brands like iPhone are very dominant in channel management. They may see end-side models as an opportunity or at least something they cannot miss. As for what they will eventually achieve, to be honest, I don't care that much; it has little to do with us.

Li Yanhong believes that these companies may be weaker in true cloud models and may eventually rely on one or more cloud-based large models. Baidu also hopes to provide cloud model capabilities for them.

"As for future innovations in edge-side models, I haven't given it much thought since I don't care that much. If it merely copies the capabilities of cloud models, it wouldn't be very exciting. It would just be smaller and less powerful. At most, people might think it offers better privacy protection because data isn't transmitted to the cloud, and that's it."

Baidu Competes on Efficiency, Not Price Wars

Robin Li stated that the current large model market is quite chaotic. There isn't a strong consensus yet on which model is the best, which offers the best value for money, whether to use public cloud or private deployments, and so on. They need to move forward a bit more. In the end, the competition won't be about price, but about efficiency.

"Whoever has the highest efficiency—who can minimize costs while achieving the same results, or maximize results with the same cost—will come out on top in future competition."