Welcome back to The One-Pager.

A quick roundup of the latest news in AI and financial modeling.

Skim it, grab something useful, get on with your day.

Your Brain on AI

Remember the "this is your brain on drugs" commercial?

That's exactly what I thought when I found this study -- researchers at MIT put brain scanners on people while they used ChatGPT to write essays and measure their brain activity.

Over the course of four months, the MIT team asked 54 adults to write a series of three essays using either AI (ChatGPT), a search engine, or their own brains (“brain-only” group). The team measured cognitive engagement by examining electrical activity in the brain and through linguistic analysis of the essays.

Two interesting punchlines emerge:

1. Perhaps not surprisingly, the ChatGPT group showed up to 55% lower neural connectivity than the brain-only group. I don't know about you, but I definitely feel myself glaze over when I'm waiting on AI to do something and/or give me some insight. So despite this being only a 54-adult study, my personal experience seems to align.
2. In the final session (session 4), the original ChatGPT users were then asked to write essays without ChatGPT, and showed weaker neural connectivity than the "brain-only" group. The researchers coined this term "cognitive debt," and in an interview the lead researcher said "there is no cognitive credit card. You cannot pay this debt off." In other words, AI creates a "brain gap" that you can't get back.

Now, as this article nicely points out, the deficit is inconclusive since the "brain-only" group had three sessions of practice while the AI group only got one unaided attempt -- so the gap might just be a reps problem, not a permanent one.

Punchline: until the "AI group" got the same number of "brain only reps" as the "brain group," the findings are inconclusive.

Like I mentioned above, I feel "AI numb" all the time, but definitely take this one with a grain of salt -- it was a small group of adults (only 54), only 18 adults did the 4th trial (AI to brain-only), and it has not been peer reviewed (but at least the authors pointed this out).

Regardless of how reliable it is, it aligns with something I’ve been saying for a while: if you skip the fundamentals and go straight to AI, you might think you're learning when you're really not, or creating a mistake you don't have the ability to recognize. And, perhaps even riskier, you might be losing a skill you may not get back.

Helpful to check out:

1. Microsoft killed free Copilot in Excel (Apr 15) – If you don’t have a paid M365 Copilot license, the in-app sidebar is gone. No announcement, just disappeared. I still haven't warmed up to Copilot even though I know it's improving and incorporates more models.
2. Claude Opus 4.7 launched (Apr 16) – Anthropic’s new flagship model. Coding benchmarks jumped and the vision upgrade is really impressive (it reads screenshots very well). Catch: the new tokenizer uses roughly 35% more tokens for the same text, so you’ll hit your limits faster than on 4.6, but I'm using it daily.
3. OpenAI acquired Hiro Finance (Apr 14) – An “AI personal CFO” startup. Second finance app acquisition in a year. This seems like a continuation of the arms race for which AI du jour will lead the finance function.
4. PwC: 20% of companies are capturing 75% of AI’s economic gains (Apr 13) – While I feel like the initial AI conversation is around productivity and cost cutting, leading companies are 2-3x more likely to use AI for growth.
5. And of course: How I’m Modeling with AI – newly released on my site, which includes downloadable Claude skills you can apply to your financial modeling right away. Don't forget though, fundamentals first! (We teach that as well)

Hope you enjoyed this short newsletter!

That's it for this one. See you in two weeks.
—Chris