After a decade-long commitment in the field of artificial intelligence, Baidu’s Ernie Bot model welcomes another significant update!
On the morning of October 17th, at the “Baidu World Conference 2023” held in Beijing, Li Yanhong, the founder, chairman, and CEO of Baidu announced the official release of Baidu’s most powerful Ernie Bot to date — “Ernie Bot 4.0”.
Li Yanhong stated that Ernie Bot has achieved a comprehensive upgrade of the foundational model, with significant improvements over the current version of Ernie Bot. At the venue, through real-time demonstrations of vague query responses, ad creativity generation, and multi-turn dialogue content creation, he showcased the strong comprehensive capabilities of the new generation of Ernie Bot.
It’s worth noting that as the foundational model’s capabilities continuously strengthen, Baidu is accelerating the promotion of AI-native application implementations. Over the past few months, Li Yanhong explained that Baidu has restructured its product line based on AI-native thinking, including search engines, Baidu Docs, Baidu Cloud Drive, Baidu Maps, and a range of other products. Baidu also launched China’s first generative business intelligence product for enterprises – Baidu GBI, which, as introduced, can significantly enhance enterprises’ data analysis and decision-making efficiency.
Li Yanhong: Ernie Bot is on par with GPT-4
At the “Baidu World Conference 2023”, Li Yanhong delivered a keynote speech on “A Step-by-Step Guide to Building AI-native Applications”, and unveiled Baidu products restructured with the large model.
The focal point was Li Yanhong’s announcement of the official release of the new Ernie Bot 4.0 and the commencement of its invitation for testing. This is a significant update following Baidu’s iteration of Ernie Bot 3.5 in May this year.
According to Li Yanhong, this is the most potent Ernie Bot to date. It has undergone a full upgrade, and its understanding, generation, logic, and memory capabilities have all notably improved over the online version of Ernie Bot, making it comparable to GPT-4.
Data shows that Ernie Bot is an industrial-level knowledge-enhanced model independently developed by Baidu. It has formed a three-layer model technology system of “foundation-task-industry”. On March 16th of this year, Ernie Bot initiated its invitation for testing. In May, the Ernie Bot model iterated to version 3.5, achieving a training speed three times faster and an inference speed over 30 times faster than before. By the end of August, Ernie Bot was made widely available to the public and, within a short time, achieved daily active users surpassing a million, ranking first in the AppStore.
The “AI Large Model Technical Ability Assessment Report, 2023”, released by the international consultancy IDC in July, showed that Ernie Bot 3.5 achieved full scores in 7 out of 12 indicators, ranking first among domestic mainstream large models. On the day of the “Baidu World Conference 2023”, Baidu’s CTO Wang Haifeng stated that compared to version 3.5, the newly released Ernie Bot 4.0 has notably enhanced understanding, generation, logic, and memory capabilities. Specifically, the improvement in understanding and generation capabilities was almost equal, while the enhancement in logic and memory was even more substantial – with logic seeing a near threefold increase and memory experiencing more than double the improvement in understanding.
At the conference, using specific application scenarios, Li Yanhong detailed the latest understanding, generation, logic, and memory capabilities of Ernie Bot. Real-time test results showed that Ernie Bot could understand “disordered descriptions, vague expression intents, and the underlying meanings in spoken words” with significant accuracy. Simultaneously, Ernie Bot was able to rapidly generate car ad posters, ad copies, and ad videos; logically provide detailed answers to complex math problems and apply learned concepts; and remember multiple character backgrounds during multi-turn dialogues in novel writing.
Baidu’s Products Fully “AI-nativized”
As significant as the basic model’s capability upgrade is the actual application implementation. In this regard, Baidu aims to realize the application value of the large model through “AI-native applications”.
Li Yanhong believes that so-called “AI-native applications” are applications developed based on the large model’s understanding, generation, logic, and memory capabilities. These abilities were not present in the past era, and they also come with a fundamental change in natural language interaction. “The more powerful the applications developed on top of the model, the more people use them, and the more productivity and work efficiency increase, the more it reflects the real power of the foundational large model,” Li Yanhong once publicly stated.
In May this year, Baidu claimed to be the first company to restructure all its products using AI-native thinking to create AI-native applications. Five months later, Baidu announced that its products, including the search engine, smart office platform, maps, cloud drive, and docs, have all been restructured with the large model.
Specifically, Li Yanhong showcased the new look of several Baidu products at the conference. For instance, when opening Baidu’s new search engine, the search results will differ from previous link displays; they will now be generated answers after the large model’s understanding and can even be presented in dynamic chart forms. Additionally, the search will recommend other related questions and achieve personalized multi-turn interactions.
Regarding Baidu Cloud Drive, it was introduced that Baidu launched “Cloud-One” intelligent assistant for individual users. Users can converse with “Cloud-One” to quickly search for files and videos, summarize knowledge points, translate documents, and other functions. Another example is the new Baidu Docs. Li Yanhong believes it’s the most comprehensively restructured product, where it understands document data to generate speeches and further auto-generates beautified PPTs, greatly enhancing productivity.
For Baidu Maps, a product with 18 years of history, Li Yanhong said that after restructuring through AI-native thinking, it has become the world’s first AI-native map product. It allows natural language dialogue to directly access functionalities and services like journey reminders, service bookings, city tours, etc., offering a “real human interaction” experience and accommodating multiple follow-up questions.
Launching China’s First Generative Business Intelligence Product
In addition to existing products, Baidu has released China’s first generative business intelligence product for corporate users based on Ernie Bot capabilities: Baidu GBI. “It’s a product built from scratch,” Li Yanhong commented at the conference.
In reality, every day in government, finance, industry, and other sectors, there’s a vast demand for data query and analysis. This data is related to the core information of these industries, such as business operations, production safety, equipment monitoring, business growth, market trends, city management, etc. Baidu GBI is said to support natural language conversational interactions to carry out data queries and analysis tasks.
Li Yanhong stated at the conference that in the field of business analysis, Baidu GBI revolutionizes traditional data analysis in three ways: firstly, while traditional BI can only be operated by professionals, GBI understands and responds to CEO-level queries in real-time; secondly, GBI offers a convenient integration method, allowing businesses to connect data and ask or analyze any data without manually navigating across databases and spreadsheets; thirdly, GBI also has a learning capability, where businesses can inject industry-specific knowledge, turning it into an industry expert.
In summary, Li Yanhong said: This product can reduce the data analysis and report writing work, which would take a business analyst several days, to just a matter of minutes.
International Investment Banks: Bullish on Baidu’s AI Monetization Capabilities
This year, the rapid rise of large models worldwide has accelerated the development process of the AI industry.
According to IDC’s forecast, by 2026, the global AI market size is expected to reach $308.28 billion (approximately 2.25 trillion RMB); China’s AI market size is expected to reach $26.44 billion (about 193.3 billion RMB) by 2026, with a compound annual growth rate of approximately 21.5% from 2023 to 2026.
In August this year, the National Internet Information Office, along with six other departments, officially implemented the “Interim Measures for the Management of Generative Artificial Intelligence Services”. The same month, including Baidu and ByteDance, the first batch of eight large models were recorded. This signifies the large model industry in China has officially entered the product implementation phase. CITIC Securities believes that the ecosystem of domestic large model products will gradually be established, services will become standardized, and the large model industry will enter the fast track of product implementation.
Regarding Baidu’s AI commercialization, Li Yanhong previously mentioned in an earnings call that “Overall, Baidu is committed to building a new growth engine based on generative AI and large language models, driving sustainable long-term growth.” Baidu’s financial results for the first half of the year showed that from January to June, the net profit attributable to Baidu (non-GAAP) reached 13.725 billion RMB, a year-on-year increase of 45.7%. Research reports indicate that last year, Baidu Smart Cloud ranked first in the domestic AI public cloud market share.
So far this year, several international investment banks have given optimistic evaluations of Baidu’s AI monetization capabilities and have upgraded Baidu’s stock ratings and target prices.
In July, Goldman Sachs reiterated its buy rating for Baidu and added Baidu to its regional strong buy list. Goldman Sachs predicts that the future general-purpose large language model in China might be dominated by a few companies, becoming the main traffic entrance, and internet giants are in a favorable position. Baidu is one of the strongest companies in the generative AI field among Chinese internet companies and could become a primary traffic entrance for future AI applications.
Goldman Sachs pointed out that according to data from the China Academy of Information and Communications Technology, the Chinese cloud computing market is expected to reach 380 billion RMB by 2023. The market will “witness” the monetization potential of Baidu’s AI cloud services and the monetization potential seen in consumer usage scenarios, such as search.
Morgan Stanley rated Baidu as “China’s Best AI Stock” in June and upgraded Baidu’s rating to overweight, further raising its target price. Its research report states that the development of artificial intelligence in China is at a turning point, the consumer market is moving from offline to online, and Baidu will be the go-to company to capture the development of the $7.4 trillion AI Internet market. Morgan Stanley’s report states that Baidu is the “most evident beneficiary” of China’s increase in AI applications, and the company has deep expertise in areas such as autonomous driving and generative AI models, as well as proprietary search data.
In fact, large models have become a high ground in tech company competitions, and the market has given them much attention and affirmation. In February, Google released its next-generation conversational AI system Bard. Although its stock performance was mediocre for a few trading days following the announcement, the market gradually realized the value of Bard, and its stock price soared (from $102/share to $142.38/share). In March, Microsoft officially announced that its Bing search engine had integrated OpenAI’s DALL·E model. The stock price had a slight decline initially, but as OpenAI’s large model capabilities deeply empowered Microsoft’s product line and Microsoft successfully transformed into an AI company, its market value surged significantly, rising from $1.79 trillion in 2022 to the current $2.5 trillion, with the stock price reaching a historic high at one point.