In October 2011, Siri made its debut as an integral feature on Apple devices, seamlessly integrated into the iPhone 4S.

Fourteen years later, the open-source AI model DeepSeek emerged as a game-changer.

In February 2024, WeChat's AI search integrated DeepSeek, blending immense traffic with an open-source model to create a powerful synergy.

While Siri and DeepSeek are both renowned intelligent AIs of their respective eras, Siri has faced criticism and even been dubbed an "artificial idiot." Unlike the closed-source Siri, open-source DeepSeek has rapidly advanced, successfully integrating into various domestic platforms and now WeChat, undoubtedly becoming a standout star in the AI world.

The Evolution of WeChat and AI Search

1. WeChat Search and AI Engine's "Past Challenges"

Incorporating a search AI engine into social software might initially seem odd. However, as a national-level application, WeChat's search function has always been a crucial entry point for users to access information. It's more convenient to complete tasks within the same app than to switch to another, aligning with the ethos of convenience in daily life.

Nevertheless, WeChat's traditional search faced issues such as fragmented information, lack of in-depth analysis, and insufficient personalization, making it difficult to meet users' evolving and diverse needs. Traditional search often required multiple clicks and filters to find complete information, with results limited to simple keyword matches. Additionally, when tackling complex problems, WeChat's traditional search lacked systematic and logical analysis, relying solely on basic keyword matching without intelligence. Therefore, addressing these traditional search issues was imperative, and with the advent of the AI era, the shift from traditional search to AI intelligent engines became a worthwhile endeavor.

2. The Rise of AI and WeChat's Exploration

In recent years, AI technology has made significant strides in natural language processing and information retrieval. Prior to integrating DeepSeek, WeChat also actively explored AI technology applications. For instance, WeChat Keyboard introduced the "One-Click AI Q&A" feature, leveraging Tencent's Hunyuan large model to provide users with a smart interactive experience. Users could input a question in the WeChat chat box, add an "=" symbol, and quickly trigger the AI Q&A function, receiving an intelligent response. However, this feature had various drawbacks, leading to a low adoption rate. It couldn't fundamentally resolve WeChat's traditional search issues and served more as a transitional solution.

With the AI era upon us, WeChat didn't lag behind.

In February 2024, WeChat officially integrated the DeepSeek R1 model, marking a new chapter in open-source AI search engines. But is this merely a temporary fix or a viable approach for WeChat and AI?

Why DeepSeek R1 Deserves Repeated Praise

1. What is DeepSeek?

DeepSeek is a Chinese AI company incubated by Huifang Quantization, specializing in developing high-performance, cost-effective large language models (LLMs). Its flagship product, DeepSeek-R1, is an open-source inference model with rapid question-answering and deep thinking capabilities, particularly excelling in math, code, and natural language reasoning tasks, such as scoring 97.3% on the MATH-500 test, surpassing OpenAI's GPT-1 model.

The model employs a Mixture of Experts (MoE) architecture and reinforcement learning techniques, supporting model distillation to train smaller models tailored to different application scenarios.

2. Open-Source Model, Free for Commercial Use

DeepSeek adheres to an open-source strategy, making model weights and training details accessible to developers and supporting free commercial use, fostering inclusive AI technology development.

After integration by mobile phone brands like Huawei, Honor, OPPO, and vivo, technology giants such as iFlytek, Microsoft, and NVIDIA also adopted DeepSeek. Now, the national social software WeChat has embarked on a journey of deep exploration with DeepSeek. So, what makes it so exceptional?

3. Core Capabilities of the R1 Model

Let's delve into the DeepSeek R1 model referenced by WeChat. Developed by DeepSeek, it's a representative of open-source large models, focusing on high-performance reasoning tasks.

DeepSeek R1 serves as a crucial technical backbone for WeChat's intelligent search engine. Its core capabilities include enhanced reasoning abilities fueled by reinforcement learning, long-chain reasoning technology, and model distillation support.

During training, DeepSeek R1 adopts a multi-stage strategy, encompassing cold start, reasoning-oriented reinforcement learning, rejection sampling and supervised fine-tuning, and full-scenario reinforcement learning stages.

Long-chain reasoning technology enables it to progressively break down complex problems and solve them through multi-step logical reasoning, demonstrating heightened efficiency. Additionally, the model allows users to utilize its output for model distillation, training smaller models to meet varying application needs. This not only reduces data collection costs but also enhances the model's generalization and self-correction abilities, mitigating the AI "hallucination" phenomenon where users receive erroneous information.

As the saying goes, "The proof of the pudding is in the eating." Since the grayscale testing of DeepSeek R1 on WeChat on February 15, what kind of user experience has it delivered? Has it met expectations?

Experience of WeChat AI Search Integrating DeepSeek R1

1. Current Status of Grayscale Testing

After integrating the DeepSeek R1 model, WeChat's AI search has significantly enhanced user experience.

Currently, WeChat's AI search function is in a limited grayscale testing phase. Some users can see the "AI Search" option in the search box, offering "Quick Q&A" and "Deep Thinking" modes.

Users who haven't encountered this feature may need to wait as it continues to roll out.

2. User Experience and Feedback

User feedback indicates that DeepSeek R1 not only retrieves information swiftly but its "Deep Thinking" mode also provides comprehensive analysis, supporting citations from official account posts and webpage information, enabling users to continue asking questions for continuous information retrieval.

While accessing information, privacy concerns inevitably arise: Will personal data be stolen or misused?

WeChat AI search prioritizes user privacy and data security, clearly stating that it won't use users' chat records and Moments data. In the era of the Internet of Everything, can personal privacy truly be safeguarded? As open-source search becomes more prevalent, avoiding telecom fraud and online scams may become another top priority.

3. Directions for Improvement

The current version still has room for improvement, such as not supporting multi-round dialogues, inability to view historical records, and longer response times for complex queries.

Future enhancements may include adding memory functions, optimizing response speed, and further bolstering multi-modal interaction capabilities.

With ongoing technological advancements, WeChat AI search is poised to play a pivotal role in more scenarios, like intelligent customer service and content creation assistance, further enhancing user experience. However, with AI's rapid development, concerns about its future impact persist. We firmly believe that AI is a tool and cannot replace humans, who possess boundless creativity.

Market Trends and Future Prospects of AI Search

1. Changes in the AI Search Market Landscape

WeChat's integration of the DeepSeek R1 model signifies a notable shift in the domestic search market landscape.

Leveraging its vast user base and high-quality content ecosystem, WeChat has further elevated the efficiency and depth of information retrieval through AI search functions.

Concurrently, the rise of emerging platforms like Xiaohongshu has disrupted the traditional search market. Xiaohongshu's search volume has surged over the past year, with its daily average search volume reaching 600 million, nearly half of Baidu's search volume.

Where does WeChat stand after integrating DeepSeek R1? Intelligence is unstoppable, and transformation and upgrading are crucial for enterprises. In the AI era, we are driven forward, and embracing AI is the key to navigating this new epoch.

2. Industry Trends and Outlook

Overall, the AI search market is in a rapid growth phase, anticipated to maintain a growth rate exceeding 20% over the next five years.

The application of AI technology in search will become more widespread, further enhancing user experience. In the future, AI search will evolve from a traditional information retrieval tool to an intelligent decision-making assistant, enabling users to fulfill complex needs like price comparisons and customized solutions through search.

WeChat AI search's development roadmap includes supporting additional functions like file uploads and continuous dialogues, further bolstering its competitiveness in the search market.

With continuous technological advancements and expanding application scenarios, AI search will play a pivotal role in numerous fields, reshaping the search ecosystem. The era of AI has dawned.