Alex Lassiter: Hello lovely people! I’m Alex Lassiter from the Minnesota Daily. You’re listening to In The Know, a podcast exclusive to the University of Minnesota.
I love Hallmark movies. There’s something very comforting about them. But I have to admit they are a bit cheap. I was binge-watching some of them a few weeks before Thanksgiving (please don’t judge me) and I started noticing how similar some of these plots were. I accidentally switched from the Hallmark Channel to another channel, and the movie I switched to was at the exact same point, narratively, as the channel I left it on. Different movies, exactly the same scene. Then a question arose. What does it take to make a Hallmark movie? And can the magic of Christmas be formed from a formula?
As a first part of unraveling this mystery, I reached out to Sarah Marsh, a theater arts and dance teacher at the University of Minnesota and a freelance actress from the Twin Cities. Marsh actually starred in a Hallmark movie. She played the main character’s sister Helen in the 2016 film. I love you always, Santa.. I wanted to know a little more about her personal experiences on set and whether the experience of being in a Hallmark movie influenced the way she views movies.
Sarah Marsh: Sometimes the atmosphere needs something a little more hopeful, a little more warm.
And I think we’ve seen that happening across the board. And I think Hallmark embodies exactly that. So, you know, sometimes when you get a script, it reads kind of corny. And what I tell my students is that you have to take it seriously and invest in it with intention. Because if you do that, it won’t come across that way.
Lassiter: After eight days of filming, Marsh was on and off. The entire movie took just two weeks, including the time Marsh was on set. It was filmed in January and released in December of the following year. Interestingly, the film was shot less than an hour south of the Minneapolis studio in Northfield, Minnesota.
Although it may sound cramped, Marsh says he had a lot of fun filming it. Even just a week was more than enough time to bring home wonderful memories from the filming set.
Marsh: You move very, very, very quickly. For example, on the first take, you nail the take, but they don’t. There was no reason to go back. On the first day of shooting, I went through a long emotional scene. My character was having a mental breakdown and it was kind of a revelation. It worked. It worked on the first take. And they were like, “Great, let’s move on.” So I said, wait, wait, “Can I try again?”
And the director came out and said: why? “And I thought, ‘Hmm, I think I could do a little better.’ It was a bold request, but as you know, I completely blew it right away on the second take. I blew it. I just blew it. I mean, I blew my lines right away. And I thought, “Oh, I shouldn’t have asked.” I shouldn’t have asked. ”
Lassiter: Hallmark movies are full of that cozy Christmas spirit, even behind the scenes. The experience Marsh had on set was wonderfully authentic, yet optimized. And companies like Hallmark, which currently releases about 40 new movies a year, can go even further by planning the plots of their Hallmark movies according to some sort of formula before filming begins. I was still wondering if it can’t be optimized. After doing a little research, I found that I wasn’t the only one who had the same question.
Three years ago, Marty Kihn, senior vice president of strategy at Salesforce, asked himself if he could combine his passion for data and love of Hallmark movies to generate entirely new plots using AI.
Marty Kinn: I thought maybe I could do this. Is a generative AI actually equipped to write a synopsis for a Hallmark Christmas movie? I was a huge fan of these Hallmark Christmas movies, so I’m just enjoying them. But they’re very formulaic, and I’ve always wondered what the formula is: is there one formula or two?
So I was interested in that question. And my end result was, can I train a computer to actually create one of these? Not the entire script, just a synopsis.
Lassiter: So how did Keene do this? By narrowing down each movie to fit into one of eight themes. Setbacks, “boss crushes,” travel mix-ups, alternate lives, major business acquisitions, rivals turned lovers, scammers, and family crises. For more information on each, original publication.
Kihn will see if language models like GPT and text generators like Markov chains can pick up on these themes and use them to generate original plots for “movie-ready” Hallmark movies. I was thinking of doing it. The results were… less than magical.
Keyin: This was before GPT 3. So generative AI, this was only two years ago, generative AI didn’t exist. You can produce sentences that sound okay, but if you make them longer than that, you’ll be watching your thoughts go through the paragraph and you won’t be able to maintain your thoughts.
Lassiter: Keene determined these eight themes by taking information from the descriptions of over 100 Hallmark movies and converting them into tokens. So names like Jack and Jill changed to and . The big company has become , and the hometown where the main characters spend Christmas has become .
By focusing on the movie’s characters, settings, and archetypes and generalizing them into placeholders, Keene essentially reverse-engineered Hallmark’s secret formula.
Marsh: I feel like I tend to be formulaic, but this is a formula that works. That’s not a negative criticism. That’s just a criticism in that it’s something I noticed. It’s a plot line and story arc that works well that people respond to.
Lassiter: Optimizing people, locations, and big story beats creates a sense of intimacy, as does the streamlined shooting schedule Marsh agreed to.
Tianxi Li, an assistant professor of statistics at the university, uses the new version of GPT in her work. Kihn’s hypothesis used GPT-2, while Li uses GPT-3 and GPT-4 for pattern recognition.
Tianshi Li: Actually, there are no predefined labels or groups you want to know about. But they have a heuristic assumption that there are a small number of classes of plots or stories that they want to formulate, and that actually creating a new plot involves changing each type a little bit.
Lassiter: What was the biggest problem with Kihn’s process?The technology of the time. But Keene’s theory that new generative AI software could officially generate fully comprehensive plots that any Hallmark executive could use right away could theoretically be within reach. There is.
Keyin: If you have good descriptions, you can actually train the model or point the model based on those descriptions to create similar but different stories based on these descriptions. You can instruct as follows.
You can then tell them the names of the characters, the setting, and even give them some guidelines. Then what comes out will probably be quite acceptable.
Lee: And the complicated part of that is how do you specify the model, the machine learning model, to do that? But the basic logic remains the same, it’s just finding patterns.
Lassiter: Whether you think it’s formulaic or free-form, there’s no denying that Hallmark works. People watch those movies. people like them. I like them. And Marsh says people need them. Regardless of how you feel about Hallmark movies, there’s an undeniably human element to them.
Marsh: There’s a struggle, there’s tension, there’s a climax, and there’s usually a positive resolution for Hallmark. And I think there’s been a lot of uncertainty in the world, especially over the last five years, and people need something to feel good about and something to feel hopeful about.
And I think Hallmark does a really great job of that, you know, they give you something hopeful and heartwarming, and sometimes we just need something warm. Yes, and it’s fun to be a part of it.
Lassiter: Of course, there are good and bad Hallmark movies, just like there are good and bad Netflix and Marvel movies. To me, a good Hallmark film is like a blanket or a fireplace. Even though it’s the same warmth, it’s definitely warmer. And you can’t get that level of warmth and intimacy from a cold, emotionless machine.
Keyin: It’s a really human perspective. I think what machines are good at is taking something that’s been done before and producing a version of it. If it’s not on the Internet, GPT can’t know it for a fact. If you’re actually creating a report, this conversation has never happened before, so I think it’s hard for a machine to generate it.
That’s one, and the other is related to emotion and empathy. This means more authentic human interaction between humans.
Marsh: They were a little more improvisational than some directors, some screenwriters, some sets, and they were totally game. It turned out to be very, very different for me and a lot of fun.
It was very spontaneous. It was very lively. It had a very strong presence. And that was really fun for me. So I didn’t mind trying it out or incorporating it, and it was a lot of fun. So it became more lifelike and quite different from other things I’ve made before.
Lassiter: Regardless of age or era, I love Hallmark movies. No matter if it’s good or bad, serious or creepy. As long as you make it with all your heart. A sense of routine and familiarity can free up concentration for writers and actors. You get more time to put the best parts of yourself into that particular movie instead of trying to reinvent the wheel with every movie. And like I love you always, Santa.highly recommend this watch. You just might find something magical.
This episode was written by Alex Lassiter and produced by Kaylee Sirovy. As always, thank you for listening. Please feel free to send us a message in your email inbox. (email protected) If you have any questions, comments, concerns, or ideas for episodes you’d like us to produce this season, please feel free to contact us. I’m Alex. This is In The Know. Everyone please be careful.
