Two Paths for AI: Expertise vs. Brute Force

3 Dec 2025 · 1 min

Watch on YouTube

0:44

AI needs either massive compute power or expert-built datasets to handle complex accounting rules. A $50 billion monthly spend might be OpenAI's path. The alternative: train models on curated data from domain specialists so they understand the difference between lease types and REIT revenue classifications without brute force.

Chapters

Oh, wow. But the key thing to keep in mind is that there's no substitute for having that expertise in it at its inception. Like, these machines aren't going to figure out the difference between a not-yet-commenced lease and an operating lease, or sublease income that should be operating or non-operating, or gains on sale that should be operating, an operating revenue for a REIT, but shouldn't be operating for an industrial company. These are all kinds of things that machines don't know how to do. Maybe they'll figure it out eventually. But I think there's 2 paths to success with AI. One is like just brute force. Like, I'm going to buy and spend $50 billion a month on compute, which is what we see OpenAI doing. The other is, am I going to be able to, to put together a dataset that can endow the machine with enough subject matter expertise to produce truly reliable insights on par with what an expert can do?