Silicon Valley Bank failed on Friday in roughly 48 hours. Watching this play out from the inside of the banking industry, what stood out was the speed and how cleanly the mechanics matched what the risk frameworks have always warned about.

Watching what rising rates have done to bank securities portfolios over the past year got me looking into the broader question. Unrealized losses on bonds are one thing. The bigger question is how rate movements affect an entire balance sheet, not just the securities line. That's what interest rate risk actually is, and I've been spending time trying to understand the framework bankers use to manage it.

I've spent the last two months on personal time exploring ChatGPT, digging into how large language models work, and running it through every banking scenario I can think of. The wonder phase is fading. What's replacing it is a more practical question. Where could this eventually help at a community bank, and where would it get a bank in trouble?

The CFPB put out a proposal last fall for what's being called open banking, and it has been on my mind since. The idea is that clients should be able to share their financial data with third parties through secure APIs, and that banks should be required to make that possible. I read most regulatory proposals trying to picture what I'll eventually have to build. This one I read as someone who has spent years building data plumbing, and from that angle it's the most interesting thing to cross my desk in a while.

There's a question that keeps coming up in every conversation about banking strategy. If most transactions happen digitally, what are branches for?

All the attention right now is on large language models. But there's a whole category of machine learning that's been quietly useful for years and doesn't require a GPU or a billion parameters. Clustering is one of those techniques. The idea is simple. Given a pile of client data, can an algorithm find natural groups without being told what to look for? I wanted to try it.

After two weeks of playing with ChatGPT, I hit a point where "wow, it's good" wasn't enough. I wanted to understand what's actually happening when I type a prompt and get a coherent response. Not at a PhD level, but enough to move past treating it like a black box.

The Fed raised rates again this week, the seventh increase this year. The federal funds rate is now 4.25 to 4.50 percent. That's 425 basis points in nine months. I've been paying more attention to the securities line on the balance sheet lately, and now I'm trying to understand what all these rate hikes actually do to that portfolio.

I wasn't prepared for ChatGPT. OpenAI released it on November 30th, it hit a million users in five days, and I signed up expecting the usual chatbot experience. Ask it something, get a mediocre response, close the tab. That is not what happened.

FTX is gone. Filed for bankruptcy on November 11, and within days the picture went from bad to catastrophic. Roughly $8 billion in client funds, not lost to a market crash or a hack, but moved. Transferred to Alameda Research, a trading firm also run by Sam Bankman-Fried, and spent on trades, investments, and who knows what else.