Скачать или смотреть How AI Models Develop Theory of Mind: Unlocking Efficient Social Reasoning

Did you know AI models use a tiny subset of their parameters to understand human beliefs and perspectives? Researchers from Stevens Institute of Technology have discovered how large language models (LLMs) perform Theory of Mind (ToM) reasoning, revealing a surprising inefficiency compared to the human brain. This breakthrough could lead to more energy-efficient AI systems. Learn how positional encoding, especially rotary positional encoding (RoPE), plays a crucial role in shaping AI's social inferences. We'll break down the key findings, explain why this matters for the future of AI, and explore how brain-inspired designs could revolutionize AI efficiency. #AI #TheoryOfMind #ArtificialIntelligence #MachineLearning #LLMs #CognitiveScience