What's this about?
Emollick surfaces two studies. One (MIT/Stanford) finds most people would be financially better off following the financial advice of LLMs (GPT-5.2 and Gemini 3 Flash), though some get better advice than others depending on how they ask. Another finds AI conversations can *increase* loneliness — contradicting earlier experiments that found reductions or ambiguous effects, with the outcome depending on the chatbot's approach.
What exactly is the catch in "better financial advice"?
The benefit isn't uniform — it depends on the *questions* people ask. A user who knows how to probe gets materially better advice than one who doesn't. It's the same "LLMs reward expertise" pattern from earlier this week, now measured in dollars.
What's the community saying?
114 + 187 likes.
- The loneliness reversal draws the most discussion — "AI makes you less lonely" was becoming conventional wisdom, and this pushes back that it depends heavily on design.
- The financial-advice result is striking but caveated — "better than most people's status quo" is a low bar, and question-quality dependence limits who benefits.
- Both feed the week's recurring theme: AI outcomes are conditional on how skillfully humans use it, not properties of the model alone.