File this under: you can’t make this stuff up (but your generative AI can).

I am prepping for a panel I am moderating next week in NYC (Pints, Prompts, and Page Builders), on the challenges (and successes) of using AI in support of content workflows.

I remembered vaguely there was a Tumblr about mistakes made by AI systems, which might provide some good visual evidence (“receipts” as the kids say) of brands who played a bit too fast and loose in leveraging AI content generation for the context of the panel.

I recalled the name was something like “There is no AI in Failure”—so I googled that, and got this AI overview:

So the Gemini AI overview is capable of spelling “failure”—listing out the individual letters in the correct order—but then continues to insist the word contains “ia not ai.” I love the “strictly speaking” caveat. It might as well have said “Well, actually . . . ”

Now, if you put quotation marks around the phrase, you get the right result: Because There’s No “AI” in “Failure” (Tumblr) via Adactio.

But it’s this kind of thing that reminds us of the fallibility of generative AI systems and how sometimes the absurdly simple thing (the number of fingers on a human hand, how many r’s are in strawberry) is what comes out wrong.

Join us at Automattic Noho for the panel, featuring:

• Abhimanyu Gupta from IEEE
• Richard Chiriboga from Smile Train
• Anne Cantera from Elementyl Intelligence

There will also be a demo of some CMS page builders (Drupal Canvas, WordPress Gutenberg, and Elementor) immediately before the panel.

Thanks to Evolving Web, Acquia, and Automattic for sponsoring. (This is part of the build up to EvolveDigital NYC in November).