Why People Are Growing Further Apart on What LLMs Can Do
Public understanding of large language model capabilities is becoming less unified, according to a discussion sparked by a post from AI researcher Andrej Karpathy. People who use the same models can come away with very different impressions: one person may see a powerful assistant, while another encounters mistakes or limitations that make the technology seem overhyped. Those judgments can depend on the tasks people try, the tools and workflows they use, and how they define success. The resulting divide makes it harder to have clear conversations about what today’s LLMs can reliably do—and where they still fall short. The topic attracted substantial attention on Hacker News, where the discussion received 279 points and 454 comments. As AI systems evolve quickly, building a shared, evidence-based picture of their strengths and weaknesses remains a challenge for users, developers, and the wider public.