MIT Sloan: LLM Financial Advice Is Good Until You Prompt Like a Normal Person
Taha Choukhmane, Weidong Lin and Matthew Akuzawa (MIT Sloan) with Tim de Silva (Stanford) tested GPT-5.2, GPT-5.6 and Gemini 3 Flash on financial advice in a paper that won the 2026 Swiss Finance Institute Outstanding Paper Award. Following the models' recommendations produced sizable savings buffers for people over 30, but the advice degraded on life-change adjustments like unemployment and on active portfolio rebalancing. The finding with the widest applicability: performance jumped substantially with structured prompts carrying explicit financial information and assumptions, because 'regular people are not writing their prompts the way a finance professor is' — a capability-versus-elicitation gap that shows up in every domain, not just finance. The story drew 310 points and 308 comments on HN.
↳ Follow the thread