Almost everyone has had the frustrating experience of listening to a parent try to recall the title of a film that they cannot, for the life of them, bring to mind. Oh, they’ve got some vague plot details, maybe they know that one actor from that other thing they were in, and they’ll no doubt remember some unhelpful context (“You know, I watched it when I came home from the dentist…”), but the name is a complete mystery. You’re not IMDb, and you’re not a psychic. How can you help?
Maybe you don’t have to — at least not by yourself. As AI enthusiasts playing around with ChatGPT and Google’s Bard have noticed, this software is useful when you’re trying to remember specific movies, songs, and books. In fact, the technology is sophisticated enough to interpret a muddled synopsis from almost any parent with reasonable accuracy.
In the example below, redditor u/Steelizard fed ChatGPT a broad description from his mother: “Help me remember a movie. It was about a city of aliens on earth with one guy in charge of the city but suddenly got attacked and started to transform into an alien.” The bot correctly matched this to the storyline of the 2009 sci-fi thriller District 9, providing its own summary. “ChatGPT got it on the first try,” u/Steelizard reported. “Bard did also get it with the same prompt but in the third draft response and among 30 other options.”
Of course, the more parameters and details you provide — the time range when the film came out, actors or directors, settings and themes — the better the odds a bot will score a direct hit. But the real test of its abilities lies in whether it can translate a half-baked memory of a film into information about that film.
So, to generate more such content for experimentation, I asked internet friends to share the funny ways their parents have tried (or failed) to recount movies they’ve seen. The first interpretation I got was an absolute gem: “You know, the one with Warren Beatty. It’s funny. He plays the [little tooting horn motion]. There’s football. Dyan Cannon and that other funny guy. You know.” Except for the gesture — which I changed to “He plays the horn” — I entered this text as written and asked ChatGPT what movie I was talking about.
“It sounds like you’re referring to the movie Heaven Can Wait (1978),” ChatGPT replied. “Warren Beatty stars in the film as a football player named Joe Pendleton, who dies prematurely but is given a chance to return to Earth in the body of another person.” Bingo. Mom didn’t even have to mention the distinctive metaphysical premise of the comedy, which would certainly have helped, but no matter. Bard correctly guessed Heaven Can Wait as well.
Still, the bots had the distinct advantage of two actor names and the football connection. How would they fare with even less information? I tried another parental synopsis from my crowdsourcing effort: “The one with Bruce Willis. You know, the one where he’s nude.” Bard, apparently offended by a hint of eroticism, replied, “As a language model, I’m not able to assist you with that.” ChatGPT bravely took a stab at it, however, coming up with The Whole Nine Yards (2000), because it contains “a memorable scene” in which Willis appears nude. Fair enough, but incorrect. “No, that’s not it. What’s another movie where Bruce Willis is nude?” I asked. ChatGPT changed course and suggested the psychological thriller The Color of Night (1994) — the correct answer!
Impressive, though I wanted to go even further. It’s probably no surprise that when I gave the bots descriptions that contained factual errors. Identifying actors indirectly (as in, calling someone “the guy from Saturday Night Live,” or calling Dustin Hoffman “The Graduate“) caused issues. If you mistakenly believe a musical was made by Jason Robert Brown and include that in your input, the bots are helpless to lead you to the right conclusion. Likewise, you won’t have much luck if Dad thinks Gattaca (1997) was about a kid who “couldn’t become a fighter pilot because of his IQ.” (It’s a dystopian sci-fi drama in which Ethan Hawke, oppressed as a genetically “inferior” man, is barred from joining a space exploration program.) And while ChatGPT had almost zero to work with when I asked for “that movie with the guy who yelled” — by which a parent meant Platoon (1986) — it gave what I found to be a more than plausible conjecture in Network (1976).
I had two more meager movie synopses to try, both from a coworker’s mother. In the first, she recalled Spider-Man: Into the Spider-Verse (2018), like this: “It has that spider New York guy, but Black. What is it?” ChatGPT, assuming “Black” referred to a costume, initially tried Spider-Man 3 (2007) but got Spider-Verse on the second try. Meanwhile, Bard nailed it. “Are you referring to Miles Morales? He is a Black Spider-Man who first appeared in the Ultimate Marvel universe in 2011,” it replied, noting that the character made his screen debut in Spider-Verse.
I thought the bots had no chance on the last query, which was simply “Blue People 2.” Yet, against all expectations, Bard and ChatGPT each knew I was talking about the Avatar franchise. The former suggested movies in the same vein as the 2009 original, and the latter — having no data beyond a September 2021 cutoff — confidently informed me that while a sequel hadn’t been released, “Avatar 2 is currently in development.” Not too shabby for a bot with out-of-date information.
All in all, then, outsourcing the task of figuring out what the hell movie your parents are trying to reconstruct is one of the more amusing niche applications of large language models. Surely it’s just a matter of time until digital assistants like Siri and Alexa gain this capability and can help mom and dad find relevant titles on streaming platforms if given just a few stray impressions of the film. That way, younger generations can focus on more important stuff — like watching movies with names we’re bound to forget someday, too.