It's not a bug, it's a probability: 'Now' vs 'Not'

You know the one. You’re firing off a message on Slack, your fingers moving faster than your brain, and you type this:

You typed: "Let's not move into the deployment.”

You meant to type: "Let's now move into the deployment."

A single three-letter word, a ghost in the machine of your tired brain, completely inverts the meaning. An instruction becomes a cancellation. A "yes" becomes a "no."

It’s a simple mistake. But it got me thinking about the flawless, logical, silicon-based "brains" we're all integrating into our systems. What happens when an LLM, a machine that doesn't make typos, faces its version of the "now" vs "not" problem?

The scary part?

It’s not a mistake for the AI.

It's the correct, logical outcome.

And if you're an engineer, that should make you pause.

To get what I mean, we need to quickly talk about "tokens"

Read full blog at unengineered - https://unengineered.substack.com/p/not-vs-now