AI transparency

The short version

Issues are written with AI assistance. I direct the work, decide what goes in, and read every word before it is published. The measurements and claims are real and checked. The illustrations are AI-generated; the diagrams are not.

Where AI is used

• Research and verification. Finding sources, pulling up current documentation, and checking that a quote or figure matches its original rather than my memory of it.
• Drafting and editing. Turning notes and findings into prose, tightening it, and cutting it down. Some passages are drafted by a model and edited by me; some are the reverse.
• Measuring. Counting things on my own machine and in my own accounts — files, invocations, costs — so that the numbers in an issue come from a command that was actually run.
• Cover illustrations. Generated with an image model and then optimised. Every cover image on this site is AI-generated unless an issue says otherwise.
• Code. This website was built with AI assistance too.

Where AI is not used

• Diagrams are not generated by an image model. They are drawn as code from the underlying numbers, which is why a chart of 795 items contains exactly 795 items. Image models cannot count and invent text that looks like labels. Anything you have to read in a figure on this site was rendered, not imagined.
• Measurements are never estimated or illustrative. If an issue says a number, that number was measured. Where a figure is a floor rather than a complete count, the issue says so in the sentence.
• Experiences are not invented. If an issue says something broke, it broke.
• Your data is never sent to a model. Subscriber addresses are not processed by AI in any way. See privacy.

What I guarantee

1. A human decides. Nothing publishes or sends itself. Every issue is read end to end and approved before it goes out.
2. Claims are checked against primary sources. Quotes go back to the talk, paper or documentation they came from — not to someone else’s summary, and not to a model’s recollection.
3. Numbers come from commands, not guesses.
4. Mistakes get corrected in public. See below.

What this cannot promise

AI systems produce fluent text that is sometimes wrong, and confident text that is sometimes invented. Human review catches most of it, not all of it. Tools and prices also change faster than a published article does.

So: treat issues as a report of what happened on one setup at one moment, and verify anything you are going to depend on. Nothing here is professional advice — see the terms.

Corrections

If something here is wrong, tell me at info@zulqarnainhaidar.com and I will fix it and say what changed. Corrections are marked in the issue rather than quietly edited in.

Labelling and regulation

This page exists to meet the transparency expectations of the EU AI Act for AI-generated content, and because a newsletter about AI tooling should be plain about its own. AI-generated imagery on this site is disclosed here and is illustrative — it does not depict real people, places or events.

Why say all this

The useful thing about this newsletter is meant to be that the numbers are real and the failures are reported honestly. That claim is worth less if I am vague about how the writing gets made. So: heavily AI-assisted, entirely human-directed, and the facts are checked.