Richard Crowley

@rcrowley.org

I just try to be useful and eat pizza. https://rcrowley.org

Me: [sips tea, calmly] Clock: [ticks to 9:00am] Slack: HEY IT'S AFTER THE HOLIDAYS HERE IS AN INDIVIDUAL NOTIFICATION OF EACH OF THE 21464863 THINGS YOU PUNTED UNTIL AFTER THE HOLIDAYS

Computers taking over writing software would be just as sad as computers taking over making art. (Programming is fun!) Fortunately, the averaging machines can’t write novel programs any more than they can make original art.

Pundits on KQED today were saying Warner Brothers Discovery has so much debt that the only sensible path is to pursue a sale. What about going out of business? They seem to be legendarily bad at it.

Two-part question: 1. Isn't the vote on ACA subsidies Mike Johnson just canceled _exactly_ the vote promised the Democrats in exchange for caving and reopening the government? 2. WHY WON'T THEY EVER FUCKING LEARN?

It’s that magical time of year when the leaves have fallen, dusk comes early, the air cools, the ground frosts overnight, and the Republicans deny climate change as if they’ve never heard of winter before.

One of my most deeply held inconsequential beliefs is that you should always plug into the bottom of a pair of outlets first. Then it’s much easier to plug in a second thing.

Something I've been wondering about LLMs: The Wolfram article and others mentioned how LLMs don't strictly emit _the most_ likely next token. This apparently makes the output more interesting and read more naturally. Does this technique send it down bad paths and cause the hallucinations?