The widening gap in shared understanding of LLM capabilities

Andrej Karpathy has highlighted a growing divergence in how different groups perceive the capabilities of Large Language Models (LLMs). According to Karpathy, there is an increasing gap in the shared understanding of what these models can actually achieve, leading to confusion and misaligned expectations. While researchers and developers working directly with the technology often have a nuanced view of its current limitations and strengths, the broader public and corporate sectors frequently oscillate between extreme skepticism and over-hyped optimism. This lack of a common baseline makes it difficult to effectively integrate AI into workflows or establish realistic regulatory frameworks. Karpathy suggests that bridging this communication divide is essential for the healthy development and deployment of future AI systems, as a shared reality is necessary for both innovation and safety.