"Not because they copied each other. Because the constraints are physics."

Why do models tend to fall back on the ‘it is not X, it is Y’ structure so frequently?

Because they were RLHFed using corpospeak text that relies on that sort of cliche.

It’s too “load bearing”

Found the "smoking gun"

Like nails on chalkboard

It really is. And why so often? It is hard to imagine this construct is that prevalent in training data.

Isn't this common on Medium and other SEO slop that has dominated the web in the past 10-15 years?

These are (maybe) the best patterns for the vendors and their purposes but that doesn’t make them the best agentic patterns.

“Sure, let me give it a go” I skeptically thought.

My skepticism was correct. It was slop.

I usually hate when all the comments say that something is slop.

... But yeah, it's slop.

Stop posting slop.

Load-bearing slop.