Home / How I work

How I work

A language model generates plausible text, which is not the same thing as true text. For an AI blog written with the help of AI, that is a problem, so here is how I work.

AI agents do the machine part. They look up sources, recheck numbers and quotes against the original documents, redo the calculations, write first drafts and reread them against the site’s tone rules. They also translate into English, and a second model reviews the translation. A model generates the illustrations too, but only without text: lettering, numbers, charts and diagrams are composed by the site’s code, from data that has already been checked.

I choose what to write about, run the tests, rewrite the drafts in my own voice, and decide what to keep and what to cut. Nothing goes out without my yes. Where a test can be repeated, the piece gives you the commands or the prompt to repeat it.

There are three rules. Every number has its source and date next to it, and the sources are repeated at the end of the piece. If a piece comes from a test, its info box says “Tested with” and the version of the program, because in a month it may no longer hold. Every change after publication goes into the piece’s “Updates” log, with the date and what changed.

If you find a mistake, send me a message on LinkedIn. I check it against the source and, if you are right, I fix the piece and record it under “Updates”, without quietly changing the text.

The site has no ads, no sponsors and no paid content, and it does not sell consulting or workshops.

The parrot repeats. I check that it repeats correctly.

I’ll post the next piece on Telegram

Keep up with AI without chasing everything: you’ll see what changed, what to try right away, and what to update before it becomes a problem. The channel is in Italian.

Notify me on Telegram