Researching An Old Sitcom With AI

Last week I debuted a new video:

It’s a deep dive into something that I wrote about a while back in this newsletter, the sitcom Til Death, which was kept on life support long after it should have been canceled so that it could reach the four-season milestone required for syndication. Since nobody was watching and nothing was at stake, the show’s writers went in some really weird directions just for the hell of it.

Most prominently, one character becomes aware that his life is a sitcom and that his girlfriend has been recast four times, which sends him into therapy. It gets more absurd from there, with random sound effects tossed in, episodes that are shown out of order – like a wedding episode that takes place before the engagement episode – and jokes about the fact that nobody is watching and the show should really have been canceled by now.

You should watch it! (My video, I mean. Not the show. Well, maybe the show. But my video first.)

How do I approach this?

Once I decided to make a video about Til Death, I knew I would have to spend time watching the show and finding clips to pull (for the purposes of commentary and criticism consistent with fair use, natch). But I had never actually seen the show before and there were more than 80 episodes of a famously bad show. Could I avoid watching them all? How would I know which episodes were worth watching, and which I should pull clips from?

I could look through IMDb to see what actors are in what episodes to figure out which ones correspond to certain plot points I was interested in, but navigating IMDb episode by episode is tedious, made extra difficult by the fact that the episodes were listed in the out-of-order manner they aired, and not in the way that makes sense if you’re looking for a certain narrative.

So, as I so often do these days, I turned to ChatGPT for help.

ChatGPT wrote a Python script for me that went through every Til Death episode page on IMDb and pulled every actor and character name and put all that information in a massive spreadsheet so I could see at a glance who was in each episode. This was exactly what I needed!

This allowed me to quickly find certain episodes. For example, the daughter on the show, Allison, was played by four different actors throughout the show’s run. So if I wanted to find the first episode where she’s played by Krysten Ritter, I could just sort by actor and find her first episode (S01E10).

Or if I wanted to find all the appearances of all four actors who played Allison, I could sort by character to group them together! That’s how I was able to find things like these three episodes that were aired out of their original order, with Allison played by a different actor in each one:

You can see in that screenshot that I also have a row for episode ratings. I found an old website that reported TV ratings information back then and even though the site had missing images and broken links, it was just functional enough for ChatGPT to write a script for me that pulled the Til Death ratings information and added it to the spreadsheet.

You can access the entire spreadsheet yourself if you’d like to poke around in it.

Chatting about the data

I wondered what other insights I could find with the help of ChatGPT. They have a feature where you can create a custom chatbot based on whatever data you provide, so I tried that out. I named my custom bot The Til Death Cast Expert and I gave it the cast spreadsheet as reference data. I thought I could have a conversation with ChatGPT and get some interesting insights! But no. The custom bot was incredibly lazy and unhelpful.

For example, I asked about any notable actors who appeared in only one episode. It listed a handful of them, like:

• Allan Oppenheimer, the voice actor who played Skeletor

• Alan Thicke, veteran TV star

• Al Ruscio, a character actor you’d probably recognize

And then it stopped, satisfied that it had given me plenty of good examples. But looking at the names, I realized it never got past the letter A in the data. So it didn’t tell me about guest appearances by Wendie Malick, Yvette Nicole Brown, Vivica A. Fox, or even Meghan Markle, which seems like a big one to overlook!

I had similarly poor results asking things that should have been easy, like “Which episode had the biggest cast?” ChatGPT said Season 2, Episode 11 had 19 cast members. I asked for a list of those actors, and it named 18 people. When I pointed that out I got this frustrating reply:

So chatting about the data turned out to be useless. It gave me either incomplete or made up information and was much harder than just looking at the spreadsheet myself. It’s an example of AI’s current shortcomings when it can’t be relied on to do a thorough and complete job of analyzing data and instead just pretends it did. ChatGPT excelled at writing me a python script to pull the data, but failed at analyzing the document generated by that script.

But I wasn’t done with ChatGPT quite yet.

Graphing the data

I wanted to visualize the information. But I’ve never been much of a spreadsheet person, so I don’t really know how to make graphs out of data. So I asked ChatGPT to make some graphs using this data. This it did quite well!

One thing that was helpful for me in understanding the show’s rollout was a graph of airdates:

I was able to see that FOX often showed two episodes a night, and one time they even had a four-episode block of Til Death episodes. And I could see that there were significant gaps between and even within seasons.

By asking ChatGPT to overlay the ratings on top of the airdates, I was able to make more sense of it and identify where network decisions affected the series:

1. That big peak in the second column of the graph is when FOX put the show after American Idol for four weeks to give it a boost. But when it moved back to its own time slot, viewers dropped again. There’s another peak in season two when they put it after American Idol again as the show came back from a mid-season break. But that boost didn’t help much either.

2. The break in mid-2008 is because they ran out of episodes for season two due to the writer’s strike. Only 15 episodes were made.

3. This is all of season three. FOX only aired seven episodes, including two days where they doubled up episodes, and the ratings were the lowest they’d been so far. That’s when FOX decided to give up on the show and shelved the rest of the episodes.

4. After season three you can see a huge gap. If you watched my video, you know that this is when FOX was ready to give up on the show but Sony stepped in and offered them a fourth season really cheap so the show could reach syndication. FOX agreed, and all the leftover episodes from Season 3 were merged with Season 4. They burned through them quickly and the show got its lowest ratings ever.

This was all very helpful for me to understand the story of Til Death but it seemed a bit too much information for the video. So I had ChatGPT make one more graph that just showed the ratings of each episode in order:

Hey, that’s much cleaner, and easy to understand at a glance. You can see the general trend of the ratings as they get lower throughout the show’s life, interrupted by a couple peaks from the American Idol boosts.

Now to animate it

But this was destined for a video, so I wanted some animation that would bring it to life — something where I could zoom in to the graph as it is drawn on screen. But I’m not an expert at motion graphics. I’m sure it’s simple for someone who know’s what they’re doing in After Effects, but that’s not me.

I tried a few clunky solutions masking parts of the graph or adding wipes, but the results looked pretty bad. I was almost ready to give up when I wondered: Could ChatGPT animate the graph for me?

I asked and it turns out, it could!

I had ChatGPT output this animation at very high resolution so I could zoom in to different points as I talk about them. It wasn’t as fancy as what a real animator could do with After Effects, but it worked.

The last piece of the puzzle

There was one more thing I did to make it easier for me to find certain episodes. I obtained caption files for all of Season 4 (the surreal season) and much of the rest of the series. This way if I wanted to find instances of, for example, references to Joely Fisher’s boobs — and there are a lot of them — I could do a text search across episodes for the word “boobs” to see where it comes up in dialogue and go right to that scene.

The Til Death Viewers’ Guide

After I posted the video, a lot of people commented that now they’re inspired to watch the show, but they only want to watch the surreal or weird episodes.

So, since I went through the trouble of figuring out where the show’s most interesting moments are, I’m providing my research notes as a sort of Viewer’s Guide that people can use to decide which episodes to watch. Some of my notes have timestamps to help.

At the time of this writing, the whole series is available to watch for free on The Roku Channel, but some episodes keep disappearing and then magically come back a few days later. I have no idea why.

Here now is the Viewer's’ Guide, just listing episodes with anything interesting in them, and with the most notable moments in bold:

SEASON ONE

S01E01 - The Pilot
Sets things up but not necessary to watch

S01E05 - The Garage Band
Appearance by Will Sasso

S01E10 - Daddy’s Girl
First appearance of Krysten Ritter as Allison

S01E13 - Fight Friend
First appearance of Margaret Cho as Nicole

S01E16 - The Italian Affair
15:40 Jessica Chastain as restaurant hostess
19:30 Cameo by Ray Romano
20:20 Brad says he “looks like someone I used to work with”

S01E17 - Clay Date
First appearance of Timm Sharp as Doug

S01E21 - Webby’s Not Happy
Appearance by Ted McGinley, aka the patron saint of shark-jumping

SEASON TWO

S02E03 - Come Out & Play
Appearance by John Cho

S02E10 - Really Big Brother
First appearance of J.B. Smoove as Kenny
18:00-ish “This is my brother. My big brother!”

S02E11 - Raisinette in the Sun
Appearance by Yvette Nicole Brown
00:47 “That’s our we’re-not-gay buffer seat”
12:56 “I’m not any good at talking. I’m not a chick.”

S02E12 - Snip/Duck
First appearance of Ken Jeong as Dr. Park

S02E13 - Sob Story
Appearance by Wendy Malick

SEASON THREE

S03E02 - Joy Ride
Appearance by future royal Meghan Markle

S03E03 - Dreamguys
Appearance by Vicki Lewis

S03E04 - Sugar Dougie
First appearance of Laura Clery as Allison

S03E06 - Circumdecision
Appearance by Alan Thicke as himself

SEASON FOUR

A good rule of thumb for Season Four is that if the episode has Kenny in it, or Allison is blonde (and played by Laura Clery), you’re watching one of the unaired episodes from Season Three that got shuffled in with Season Four. If you’re just looking for the surreal storyline, you can skip those.

S04E01 - Doug and Ally Return
First appearance of Lindsey Broad as Allison
Joke about how she looks like a whole new person
Joke about “sandy cracks” at a nude beach
16:24 Joke about living in a sitcom
18:43 Joke about a word you can’t say on network television

S04E02 - Separate Beds
First appearance of Kathleen Rose Perkins as Ms. Duffy
Possible reference to classic TV sitcoms: Eddie and Joy sleep in separate beds this episode

S04E03 - Eddie’s Book
First appearance of Barry Bostwick as Doug’s dad
First appearance of Richard Lewis as Miles
Boob/zeppelins joke in the opening
21:00 Random BOING sound

S04E05 - The Courtship of Eddie’s Parents
Appearance by Valerie Harper as Eddie’s mom
Appearance by Jerry Adler as Eddie’s dad*
Appearance by Vivica A. Fox
*Fun fact: Joely Fisher’s dad is named Eddie

S04E07 - The Buffer
Episode originally for Season 3 that explains why Kenny moved in
First appearance of Susan Yeagley as Simona according to IMDb, but I could not find her in this episode

S04E08 - Joy’s Out of Work
02:08 “You forgot to put the C in Canal” joke
That scene ends with random BOING sound
07:30 “Naughty pine / Knotty pine” joke
07:50 “Could you pick up my woodpecker?” joke and another random BOING sound
10:09 A vodka joke that might be a reference to the movie Airplane
15:49 Oral sex joke “I wouldn’t spit, I swallowed.”
21:00 Random ELEPHANT sound

S04E09 - Hi Def TV
First episode where Doug questions the nature of his existence
02:44 “I’m not sure I exist”
03:29 “What if we’re all characters in a sitcom?”
08:06 “If this were a sitcom, it would have been canceled a long time ago”
Doug asks why there are only generic products
09:04 “I saw this on Blossom”
09:37 Doug can’t swear
10:00 “A hot chick like you would never go for a guy like me in real life”
10:12 First appearance of a BOOM MIC
12:30 They watch American Idol. Possible reference to the fact that the show used America Idol to help boost its ratings?
17:40 Doug looks right into the camera and says, “Hello?”
18:00 BOING sound

S04E10 - The Not-So-Perfect Couple
First time we see Kevin Nealon and Susan Yeagley as a couple
This is confusing since the episodes aired out of order. We see them here as though Eddie and Joy already know them but the episode where they were introduced airs later as S04E14.

S04E11 - Independent Action
05:00 “Still think we’re living in a sitcom?”
14:13 “Stop all this craziness about being in a sitcom!”
15:07 Doug ends a scene with “All right, that’s lunch!”

S04E13 - The Break-Up
First appearance of Mayim Bialik as Dr. Bialik
01:30 “Striking the set!”
02:11 BOOM MIC
05:49 “There’s no ceiling”
06:10 “I’m worried we’re gonna get canceled”
06:25 BOOM MIC
06:40 Doug forgets a line and shouts “Line!” which is then read to him by someone off camera and he replies “Thanks, Lizzie.”
09:37 We meet Dr. Bialik
10:10 “I am the girl who played Blossom”
11:52 “My girlfriend has looked like 3 or 4 different people”
12:19 “I can assure you, nobody is watching. It’s like Friday night on FOX”
13:20 “I’m gonna give you a catchphrase”
18:32 He talks to camera again “big laugh for old time’s sake”

S04E14 - The Perfect Couple
First episode with Kevin Nealon and Susan Yeagley in as-originally-intended order. This is the one where they meet Eddie and Joy.

S04E15 - The Check-Up
09:00 “None of this business about saying you’re living in a sitcom.”
“In a sitcom the neighbors are never smarter than the leads".”
14:15 Random SAWING sound

S04E17 - Check Mate
Features Kate Micucchi as a waitress (she later plays Allison)
09:43 Windows Startup sound? Similar to tada.wav

S04E18 - The Concert
First appearance of Martin Mull as Whitey
2:08 Doug says “What a coincidence that what you were just talking about turns out to be the plot of this week’s episode!”

S04E19 - Merit Pay
First appearance of Kate Micucci as Allison
1:39 Doug reacts to new Allison: “Who are you?”
3:45 “Too bad we can’t recast him”
10:50 Returns to Dr. Bialik
11:47 “My wife is being played by a complete different actress”
12:28 “We all know that Roseanne’s not crazy”
19:42 This new Allison loves the stuff she didn’t used to, so Doug accepts her as the new Allison
This episode includes lots of random sounds like a GONG, ELEPHANT, etc.

S04E20 - The New Neighbors
First appearance of Gilbert Gottfried as Tommy
15:28 A little musical number

S04E21 - The Wedding
This is the animated episode, also the wedding episode
05:47 George W Bush gets waterboarded
09:51 “Lady Marmalade” musical number
12:15 One of many references in this episode to Joy’s boobs
14:40 “Jumped the shark” joke

S04E22 - Ally Abroad
In this episode, Doug and Allison get engaged. It airs after the one where they get married.

S04E29 - Big Man Little Man
This episode is all about penis size
01:35 “When you fly, do you have to buy an extra ticket for that thing?”
02:12 “It should be hanging in the window of the Carnegie deli”
04:18 Random BOING sound effect
09:28 boob jokes
This episode also has a weird S&M Dream Sequence

S04E30 - Brother’s Keeper
Appearance by Richard Kind as Charlie

S04E31 - Work Wife
18:40 Fake freeze frame with text crawl in Comic Sans
(Reminiscent of a fake freeze-frame on Roseanne)

S04E32 - Baby Steps
Appearances by Mayim Biyalik’s Blossom costars Michael Stoyanov and Jenna von Oÿ as themselves, plus a guy who thinks he was Joey on Blossom
“What about the Woodcocks?”
“When we’re not on camera, none of us exist”
04:55 Random FART sound
10:20 “We’ve had a lot of time slots”
“22 minutes, we gotta wrap things up”

S04E33 - Let’s Go
Appearance by M. Emmet Walsh
Appearance by Ron Zimmerman as gun-carrying teacher who makes a tasteless joke about teachers not going through metal detectors
01:45 Random CRASH sound
03:00 Luther is hiding amongst the garden gnomes for no reason
08:15 bad joke about teachers and guns
16:07 Random FART sound
17:08 Another boob joke
19:20 Random RATTLE sound

S04E34 - The Baby
Intended ending of the series
Followed by 3 leftover episodes from season 3 that never aired

Look, people. This is what I do. If I hear about an interesting-but-terrible little-known sitcom, I’m not just going to mention it in passing. I’m going to utilize the latest state of the art artificial intelligence in my efforts to bring the story to life. If you say you want to watch that interesting-but-terrible little-known sitcom, I’ll give you all my notes so you can watch just the best worst episodes. All I ask in return is that you share my little newsletter with someone who you think will enjoy it.

And, if you can, maybe buy me a coffee or consider becoming a paid subscriber!

And that’s it for another newsletter! Thanks as always for reading. See you next time!

David