Seznam’s AI Leap: How SeLLMa is Revolutionizing Search in the Czech Republic
The Czech Republic’s leading search engine, Seznam, has taken a bold step into the future of search by integrating artificial intelligence directly into its platform with SeLLMa, a large language model specifically designed for the Czech language. This innovative AI is woven throughout various Seznam products, marking a significant departure from traditional search engine functionalities.
SeLLMa empowers Seznam to answer complex queries that go beyond simple factual requests. While straightforward questions like "Who is the president of XYZ?" or "What’s the weather like in XYZ?" can be easily handled, SeLLMa truly shines when tackling intricate questions, synthesizing data from multiple sources to provide comprehensive answers. These answers are prominently displayed in a colored box, either at the top of the results page or right after any advertisements. Notably, users can even rate the AI’s response, providing valuable feedback to Seznam for continuous improvement.
However, Seznam’s AI-driven search isn’t all-encompassing. Currently, SeLLMa’s assistance isn’t guaranteed for every search query. The system lacks a method to proactively identify which queries would benefit most from AI analysis. Users can’t force SeLLMa to provide an answer either. This selective approach is likely a strategic decision to manage resource consumption, as processing complex queries with AI can be computationally intensive. Despite its limited application at present (handling approximately 1% of search engine traffic), Seznam is actively working on expanding its AI capabilities.
SeLLMa’s influence extends far beyond search results. The AI is employed to generate concise news summaries, analyze user sentiment and topics within online discussions, and even explains why certain search terms are trending.
Seznam Opens Up: Making SeLLMa Accessible to Developers
In an exclusive interview with Archyde, Leon Padera, CTO of Seznam.cz, delved deeper into Seznam’s ambitious AI strategy.
Archyde: Hello, Leon! Your integration of AI with SeLLMa has certainly generated significant buzz. What inspired the creation of this groundbreaking technology?
Leon Padera: Thank you for having me. The idea behind SeLLMa was rooted in our desire to create a more intuitive and human-like search experience for our users. We had been exploring AI integration for some time, and when we witnessed the potential of large language models, we decided to develop a tool tailored to the Czech language. That’s where SeLLMa came to life.
Archyde: You mentioned that SeLLMa currently handles about 1% of Seznam’s search traffic. Can you elaborate on the initial challenges faced in implementing AI for search queries, and how you’re gradually expanding its capabilities?
Leon Padera: The initial challenge was precisely identifying which queries would benefit most from AI analysis. We didn’t want to overwhelm the system with every query, considering the computational demands of processing complex queries with AI. Currently, SeLLMa is activated for approximately 1% of our traffic, but we’re constantly working on refining the system to proactively identify queries that would benefit from AI assistance. Our goal is to gradually increase SeLLMa’s reach without compromising overall performance.
Archyde: SeLLMa provides direct answers to complex queries, displayed prominently on the search results page. How does SeLLMa manage to synthesize data from various sources and generate such detailed responses?
Leon Padera:
SeLLMa is trained on a vast and diverse range of data sources, encompassing structured databases as well as unstructured text found on the web. This broad understanding of data enables SeLLMa to synthesize information from various sources and present comprehensive answers. Our team is continuously working on refining SeLLMa’s models to ensure accuracy and relevance over time.
Archyde: Beyond search results, Seznam is utilizing SeLLMa for news summaries, sentiment analysis, and trend explanations. Are there any exciting upcoming features you can share?
Leon Padera:
We’re continuously exploring innovative ways to integrate AI into our platforms. One highly anticipated feature is SeLLMa’s ability to understand and generate context-specific search suggestions. For example, if someone searches for a recipe, SeLLMa will understand they might be looking for ingredients or cooking tips and provide relevant suggestions. We’re also actively working on making SeLLMa’s capabilities available to third-party developers through an API.