How to use an AI assistant for investing research

Ask what a variable interest entity is, why an inverted yield curve worries people, or what a company's filing means when it says it took a goodwill impairment, and a good model will explain it better and more patiently than most humans will.

A language model predicts plausible text. A specific revenue figure for a specific quarter is exactly the kind of thing it can produce fluently and wrongly, and it will not sound any less confident when it does.

Any number an assistant gives you should come with a link to where it came from. If it cannot produce one, treat the number as a rumour. This single rule removes most of the risk of using AI for research.

SteadyShares's advisor is wired to the same data the rest of the site is: the filings, the statements and the valuation model, so what it tells you about a company is what the company reported, and you can open the page and check it.

The best use is not asking what to buy. It is asking what would have to be true for your thesis to be wrong, and then making it argue that case as well as it can. A model has no ego invested in your position, which makes it a better sceptic than most friends.