ChatGPT for Finance: Promise and Peril

Over the past month or so, my inbox and DMs have been flooded with questions about AI, in no small part due to my interview with Professor Stuart Russell at Berkeley and our recent webinar with ROBO Global’s Zeno Mercer and Resolve Asset Management’s Adam Butler. One crystal-clear pattern has emerged. Everyone wants real-world, specific examples of how AI can and can’t enter into an advisor’s, investor’s, or creator’s workflow.

I aim to please. Today, I’m going to peel back the curtain and give some concrete examples of how to use ChatGPT effectively to accelerate investment research and content creation. Meanwhile, I’ll also dispel some of the hyperbole around what ChatGPT can and can’t do.


What is ChatGPT?

ChatGPT is a large language model (LLM). In plain talk, that means a bunch of nerds ran a computer program to scrape all the language in the world they could find (e.g., the Internet), then built a statistical model of the relationships between words. For example, the words “blue” and “color” show up next to each other often, so the connection between those two words gets a big weight. But “blue” and “apple” do not, so they get a lower weight.

An LLM makes these connections not just between word pairs but between huge blocks of words all in relationship to each other. Then the model is fine-tuned through a ton of human work and input until the model is released.

Once in the wild, the LLM can be asked to do work (“prompted”) based on its understanding of language. I emphasize that phrasing because really, that’s all “it” is: An incredibly powerful language modeler. Many “LMMs For Dummies”-style guides exist, but here’s a good one I like that splits the difference between “no math” and “code” in its explanation: