Anthropic's Andrej Karpathy says "the external world has seen the preview" of what AI can actually do
decoder ·
Andrej Karpathy, a co-founder of OpenAI and current Anthropic researcher, highlights a significant disparity in public perception versus the actual capabilities of advanced AI systems. He categorizes the global population into four tiers, noting that approximately 75% have barely interacted with large language models, while casual users view them merely as improved search engines or writing aids. In contrast, about 20 million professionals in mathematics and programming are experiencing substantial productivity gains, completing complex projects in days rather than weeks. A tiny elite of roughly 5,000 individuals has access to internal frontier systems capable of coordinating swarms of agents for tasks like cybersecurity and scientific discovery. Karpathy describes this situation as "the weirdness of the moment," where most people see only a modest uplift while a small group faces vertigo from vertical capability curves. He attributes this rapid advancement to reinforcement learning focused on digital, verifiable, and economically valuable domains like coding. Meanwhile, human comprehension and review processes are struggling to keep pace with these accelerating technological developments.
decoder