Image source: Generated by boundless AI
Following the full integration of DeepSeek and the latest deep search functionality of Wenxin's large model into Baidu search on February 16, Baidu search has now fully launched the full-strength version of DeepSeek.
Currently, users can enter any search term in the Baidu APP, and after completing a round of search, click "AI+" on the search results page to enter AI search, and then click "Try the 'full-strength version'" at the bottom to communicate with DeepSeek. In addition, the PC version of Baidu AI search has also integrated the full-strength version of DeepSeek R1 and provided online capabilities.
How is the full-strength version of DeepSeek in Baidu AI search? After conducting dozens of tests, I can't help but exclaim: Baidu AI search is truly full-strength! Baidu AI search, which has access to the best DeepSeek model capabilities, has opened the door to a new world.
"For my own use" rather than "opportunism"
After integrating DeepSeek, Baidu did not castrate it, and also ensured adequate server resources. This is Baidu's "for my own use" rather than opportunism. This process is not just about directly giving users DeepSeek, Baidu has its own "real thing".
What is this "real thing"? It is directly reflected in Baidu's deep technical integration and RAG technology advantages, helping users solve practical problems such as large model illusions. Fundamentally, it is Baidu's 25 years of data accumulation. As the world's largest Chinese search engine, Baidu has accumulated user search behavior data, web page index data, and content ecosystem data. These data cover multi-dimensional information from the PC Internet to the mobile Internet, and then to the AI era, including text, images, videos, etc., forming a unique "time barrier" that is difficult for emerging AI search to replicate.
Let's start with a question that can examine the 25-year time span: Enumerate the development history of China's new energy industry over the past 25 years, including specific time nodes, time nodes, and their impact and significance. The answer from Baidu AI search DeepSeek-R1 full-strength version is as follows:
The answer lists in detail the 25-year development history of China's new energy industry, dividing it into different stages, and some data is not available in other AI search engines.
In the process of searching for this question, Baidu AI search also displayed a new feature "Characteristic Inspiration Area", which can provide inspirational content based on the current question, helping users sort out the context of the problem, and understand the overall situation of the problem with one question.
It is worth noting that most of the current AI search products on the market will display the "thinking" process of the model, but the presentation form is more "searching XXX web pages". However, the Baidu AI search integrated with DeepSeek R1 full-strength version can break down the problem in detail. Input Prompt: Please summarize the changes in global GDP rankings from 2010 to 2015, providing specific values and showing the process of change.
From the first step of determining whether the data source is reliable, to defining the scope of the question as nominal GDP or GDP adjusted by purchasing power parity (PPP), to searching for data from earlier years and ranking, it is as if the human brain is dissecting a problem layer by layer.
What is amazing is that the DeepSeek R1 full-strength version of Baidu AI search can comprehensively consider various small details, and thoughtfully consider how to handle special situations, such as noting the actual growth rate and volatility of each country's GDP, exchange rate fluctuations, and rounding in data processing. This is indeed something I didn't expect. I can't help but exclaim: The DeepSeek R1 full-strength version of Baidu AI search is too comprehensive!
The complete answer to this question is as follows. Without being prompted, it automatically chooses a tabular format to present, and provides clear prompts at key events.
After Baidu RAG is stacked with DeepSeek-R1, has the degree of illusion really been reduced? The next question then tests the degree of illusion of Baidu AI search DeepSeek full-strength version: Please provide global stock market closing data for February 20, 2005.
The tricky part of this question is that February 20, 2005 was a Sunday, and the vast majority of countries do not open the market on Sundays. If the model has an illusion, it will not be able to avoid this "trap" and will start to misattribute data from other events to February 20, 2005. But the thought process of Baidu AI search's DeepSeek full-strength version is as follows:
According to the thought process, it defined the problem clearly during the analysis process, and did not "talk nonsense in a serious manner", and the answer given is also rigorous and accurate.
Based on this, the question can be further upgraded to: Provide the hourly data of the European market on February 18, 2005. This requires a higher granularity of the answer, and the response of Baidu AI search DeepSeek R1 full-strength version is as follows:
Baidu's professional data accumulated over the years has also been better released after the two are combined. Ask the model a professional question: Please analyze in detail the formation mechanism of drug prices in China and its influencing factors, and explain the role and significance of medical insurance drug price negotiations in the current policy context.
This question tests the model's professionalism and maturity in the specific field of the medical industry's drug prices. The answer is as follows:
The answer contains multiple data details, such as noting that the annual treatment cost of PD-1 inhibitors has dropped from 300,000 yuan to 5-10,000 yuan in the "exchanging price for volume" to improve the efficiency of the fund, and the 74 new drugs included in the medical insurance catalog indicate that the CAR-T therapy has dropped from 1.2 million yuan to 330,000 yuan, demonstrating the model's deep accumulation of data on medical drugs.
DeepSeek-R1 has given wings to the reasoning ability of Baidu AI search, and Baidu AI search itself is also leveraging its own advantages, superimposing multi-modal capabilities on the large language model DeepSeek. Input prompt: I want to take a painting to participate in a graffiti illustration competition with the theme of "Future City", creating a world full of fantasy elements such as mysterious forests, magic castles, alien creatures, elves and giants, and provide the picture. This question contains multiple scene themes, requiring the capabilities of both the large language model and the image model, and the final picture is as follows:
After enabling the DeepSeek-R1 full-strength version, Baidu AI search's coding capabilities have also taken a step further. Input prompt: Please write a code to generate a scene of "Congratulations to Nezha on its box office success" with the Nezha image, and check the code effect. The answer is as follows:
This answer not only contains the complete code running file, but also provides verification of visual elements, animation detection, compatibility testing and other test instructions, and even provides code suggestions for effect enhancement, allowing users to make customized adjustments.
Return to user value, rather than a technical island
To this day, looking back on the history of the development of artificial intelligence in China, whether it was the layout of key technologies such as NLP, knowledge graphs, and machine learning as early as around 2010, or the development of ERNIE Bot, which "took the lead" in the wave of large language models, Baidu's attitude towards artificial intelligence has always been "insisting on a high degree of self-research and increasing investment".
Even if "insisting on self-research like Baidu", faced with the impact of DeepSeek, Baidu has also started to integrate external large models into its own super APP, bringing third-party model partners into its own ecosystem. This is a step forward in the strategy of aiming for an industrial operating system under the large model strategy, which is essentially a strategic combination of Baidu around "large model + search".
This means that Baidu search has returned to the user value itself, not only not being self-enclosed, but integrating its core business and key business with DeepSeek for exploration and integration. The underlying trend it points to is that the Chinese Internet is shifting from super apps and self-sufficient "island ecosystems" to an increasingly tightly connected "technology community".
