AI models cannot agree on how AI will change your job

Researchers have been using large language models to guess how AI will affect different jobs. It turns out these guesses are highly unstable. Different models give wildly different answers for the same job, making it hard to trust any conclusions.

Economists and policymakers have relied on these AI-generated scores to predict which jobs are most at risk from automation or most likely to benefit from AI tools. This paper shows those predictions are built on shaky ground. It means any policy or investment decisions based on these numbers might be fundamentally flawed.