It descended into glitch-ass slop towards the end, which I found funny. Very telling of an LLM/VLM since OCR would print straight garbage if it can't map a glyph to text.

This is actually a great example of why domain knowledge is important.

The printed ad is much easier to read despite the text being more densely-packed. This is because the LLM extraction stripped formatting (including the bolded and italicized text that directs readers towards interesting factoids) and used a system font and size (which is inconsistent and, often times, harder to read in column form) while the ad used a print appropriate serif that is consistent and easy to read on paper.

I'd like to think that this is graphic design 101, but when LLMs are threatening creative jobs en masse...not great.

But not to worry! All of the LLMs will nail this tomorrow after the ad's been RLHF'ed appropriately. minitruth doesn't sleep!