The curator's reviewGarfield strips re-captioned by a Markov chain trained on years of Garfield dialogue. Statistically, it is Garfield. Spiritually, it is a transmission from somewhere lasagna cannot reach.
Garkov is the elder statesman of this collection: Josh Millard, 2008, gluing a Markov chain trained on years of Garfield dialogue into empty Garfield strips. No neural anything - just probability tables over word sequences, the duct tape and string of generative text - which is exactly why it belongs here. This is what the monkeys typed before they got the good typewriters.
A Markov chain knows only which words tend to follow which words, with no memory beyond a few tokens. Applied to Garfield - a strip already built from a tiny vocabulary of lasagna, Mondays, and contempt - the output lands in an uncanny sweet spot: statistically it IS Garfield, rhythmically it is almost Garfield, and semantically it is a transmission from somewhere lasagna cannot reach. Jon says things like AND HE'LL PROBABLY SPRAIN SOMETHING AND, and the thought simply ends, and somehow the strip still scans.
The site is pure 2008 web - banana-yellow page, reload for more, links to the extended Garfield-remix universe (Garfield Minus Garfield, Lasagna Cat) that formed one of the internet's first great remix scenes around a corporate comic strip.
Reading Garkov after using modern models is a useful vertigo: the gap between it and Infinite Craft is seventeen years, and the entire technology of this collection lives inside that gap. The strips are wrong in exactly the way dreams about your own house are wrong. Jon has said nothing, and yet too much. An honorary proto-monkey, and the collection's roots.
