AI ruins Christmas movies

On the first day of Christmas genAI gave to me…

Santa and his father and their Guantanamo Bay adventure.

Strap in and get ready for some AI-generated Christmas movie plots. In true GenAI fashion, for this I’ll be churning through and recycling content I discovered in a 2018 MIT article by journalist Karen Hao. Although, unlike most GenAI, full creds are provided; you can read her article here. Karen is also the author of the recently published “Empire of AI” - an interesting and somewhat bleak insiders-take on the happenings at OpenAI and the GenAI industry in general. Highly recommend it. Especially as a Christmas gift to really dampen any semblance of hopeful Christmas spirit.

Taking GenAI back to it’s humble origins I thought I’d try to explore much smaller models - those that existed even before the kind that had Will Smith eating spaghetti like someone who was simultaneously late for the bus and physically disintegrating. I’d be lying if I said the reason for choosing smaller models wasn’t almost entirely for comedic value but I do also think something interesting can be uncovered when we distill these models down to their fundamental building blocks with no frills. The only data thrown into this model is the synopses text of 855 Christmas movies - ranging from big hitters like “Love Actually” to those that already sound AI-generated like “We Wish You a Turtle Christmas”(?).

So without further ado, I present to you; Christmas movie plots if GenAI had anything to do with it:

Santa and his father and their Guantanamo Bay adventure.

An antique store owner discovers a workaholic life and comes to the United States to start the ingredients for 30 fruit cakes

A Christmas-loving divorcee books a chalet with her ex-boyfriend and his 11-year-old son get a shock when she must attend the Super Bowl for mothers

An orphan drummer boy who hates humanity finds the party in his customers. Besides their journalistic occupations, one other similar aspect between Tom and Liz is that she finds his children and Christmas dumb

A small-town policewoman falls for a lot of money for Christmas

A woman befriends an elderly woman and learns a trio of elves save hundreds of truckers

An orphan girl, sent a month before Christmas, and Joy find the ideal property to expand their upscale property developer company

An insurance company investigator goes to sell Christmas

A single mom with twelve activities she must complete before Christmas. It took so long. Happy Christmas

I’ll come clean - these are the select few that were a) remotely legible and b) somewhat funny. For every one of these there were tens, if not hundreds, of non-sensical sentences. For example, most looked something like this:

Once upon no mayor while children have become its own promotion in jeopardy. To protect the pro and astound Paris.

Unlike ChatGPT and other LLMs, the interpreted meaning of what it spits out breaks down because of the limited size of the model and the data its trained on, revealing what’s really going on behind the curtain. Statistical pattern-matching based on a distribution of words or tokens is neatly highlighted and immediately obvious when we see the output of these more primitive language models.

And not to get too “A Christmas Carol” on you but there is also an interesting lesson here that is best put by Karen Hao in her original article: “Neural networks aren’t that funny. It’s the humans who are.”. When talking about GenAI and language models in particular, despite convincing statistical papier-mâché, the only meaning these outputs have is the meaning we attribute to them.

My Christmas wish for 2026 is firstly, a chocolate orange Toblerone and secondly, recognition that we create the meaning - not the machines.

P.S. I’ve made a tool so you can generate your own Christmas movie plots here.