Sikka Study Proves LLMs Fail Complex Tasks

Vishal Sikka and Varin Sikka published a paper claiming a mathematical proof that large language models cannot carry out computational and agentic tasks beyond a certain complexity, and Wired recently surfaced the work after its initial publication. The paper argues some prompts require computations exceeding model capacity, causing failures or incorrect outputs, which limits prospects for fully autonomous agentic AI and challenges claims that scaling alone achieves general intelligence.