Attention mechanism and summarization enable LLM-powered search breakthroughs.
Large language models are trained on vast amounts of text, and the underlying innovation is the attention mechanism. Not every word is important, so models learn to extract the key ideas for a given context. This ability, combined with their ability to complete the next best word, makes them excellent at summarization. That summarization capability is exactly what makes them powerful for search applications.
— 2025: The Year Of Search But Not The End Of Google · Forbes