As an early adopter of GPT, I've been following this closely. [See my 2022 blog: https://magnifina.com/articles/using-gpt-3-for-investing-and-financial-advice/]. ChatGPT is trained on data samples of published writing, so its advice is rooted in the most common ideas, not necessarily the best. In essence, it tells you what it thinks most people would like to hear.

In general, the advice suggested is not bad advice, per se, but it also isn't particularly clever, nuanced, or personalized for any individual client. For example, AI models commonly recommend maintaining an emergency fund for 6-12 months of expenses. The idea of an emergency fund is somewhat outdated, given today's ample liquidity and speed of electronic transfers. We often instead advise maintaining a general purpose investment account, which could be used for emergencies. But because of such a strong precedent, emergency funds remain frequently included in general financial advice. Interestingly, this effect isn't limited to AI. I've seen plenty of pre-AI articles with the same advice. Many humans think the same way that AI does, by recompiling what they've read most often.