Notes on engineering and AI systems

Under the Hood: What Actually Happens When an LLM Speaks

A continuation. My last post asked whether we trust AI too much. This one goes a level deeper — into the mechanics that produce those convincing outputs, and what they reveal about language itself.

read post

We Gave Autocomplete the Keys

We're starting to treat AI like it understands what it's doing, but it doesn't. It's just predicting what comes next, without any real awareness or judgment. The real risk isn't that it makes mistakes, it's that those mistakes look convincing enough for us to trust them.

read post