In his captivating DLD Munich 2025 talk, Shravan Nageswaran, CEO and co-founder of Atman Labs, argues that Large Language Models (LLMs) are not the path to superintelligence and presents a groundbreaking alternative inspired by human reasoning.
In the face of ambiguity, LLMs like ChatGPT or Claude like “struggle to form strategies to reach a goal”, Nageswaran observes. “That’s what makes them unreasonable.”
The solution Atman Labs developed brings together two historically separate AI approaches: knowledge graphs and reinforcement learning.
“Knowledge graphs are data structures that allow us to literally form this map of knowledge between concepts from any data”, Nageswaran explains.
The breakthrough came when they combined this idea with reinforcement learning: “The question we asked ourselves was, what if we could use reinforcement learning not just to explore a fixed game board like chess or Go, but to explore a knowledge graph – taking actions to form new paths in its world model until it reaches a goal?”
Watch the video to find out more about this innovative technology, why truly intelligent systems need more than pattern recognition, and how reasoning engines could help address some of humanity’s most pressing problems.
