Former Intel engineer pours cold water on AI hype: Solving math problems doesn't mean it can replace white-collar workers.

Beating AI News Flash: Jay Kruer, an engineer with experience in LLM training and CPU formal verification, stated in a post that OpenAI's use of tens of thousands of Agents to conquer the Navier-Stokes problem does not mean AI can already independently replace knowledge workers.

His core reasoning is simple: math problems have definite answers and can also be automatically checked with Lean. In real-world programming, research, and enterprise work, goals are often vague, and there is no program that can directly determine whether the result is correct.

Without reliable automated checks, humans have to keep watching AI constantly. Writing task rules strictly enough also requires expensive domain experts. Kruer believes that this part of the supervision and acceptance cost may even be higher than doing the work yourself.

He therefore calls today's LLMs "cheating interns": very capable and fast at work, but not yet safe to let them manage things on their own.