I asked the machine to explain itself

PLACEHOLDER: dummy copy written to show the tone and layout. Not a real post, not a real experience. Replace or delete.

There is a particular kind of certainty you only get from a machine that has never once been embarrassed.

Ask a large language model a question and it answers the way a man at a wedding explains the economy: fluently, at length, and with no visible doubt. Ask it again, slightly differently, and you may get a completely different answer delivered with exactly the same smile.

This is, I am told, a feature. It is predicting what a plausible answer looks like. Plausible is doing a lot of heavy lifting in that sentence.

So the plan is simple. Treat it like a very keen colleague who has read everything and checked nothing. Be nice. Verify the bits that matter. Say “are you sure?” and then watch it apologise and change its mind. Which is either reassuring or the opposite.

More on that next time. I have to go and fact-check a robot.

AI up me duck