Est. Read Time: 7 minutes. Read Time brought to you once again by the Ashburton Energy + Hair Logistics Group, in association with the Bradley Hills Bureau of Corrections + Housing.
New to the SternJourn? Check out the best of 2022, 2021, or 2020.
Shows
BURBANK, Friday April 7th @ 9:30pm. Flappers Yoo-Hoo Room!
Well Sternal Journalists,
It’s finally here. I’m using GPT-4 to write a SternJourn. Someone, if they cared at all, could have a lot of reactions to this, but let’s assume that two of the most common ones would be 1) What is GPT-4 and 2) Isn’t that cheating?
The easier one first:
What is GPT-4?
GPT-4 is one of the big AI things everyone is talking about. It’s the one OpenAI created. As I understand it, GPT-4 is pretty much the reason people are talking about AI so much right now. If that doesn’t make sense, let me see if I can enlist GPT-4’s help. Here’s an exchange I just had with it while trying to explain what it was:
To be clear, all of you are very smart (it is one of the benefits of having such a *ahem* selective subscribership that I can actually almost know that for certain). But I did want to get it across as simply as possible. And I think it did! I think it’s even more simple to just see how that exchange happened and understand that none of its response is pre-loaded, it wasn’t trained to answer my specific question, but rather every potential question.
It was released two weeks ago, but requires a $20/month OpenAI subscription to actually use so I don’t know many people who have actually messed around with it. I only did because, in the same 24 hour span, (a) a friend sent me this article about how there are well-paying jobs sprouting up in the industry of “AI whispering,” or basically figuring out the right things to ask an AI to get it to write the thing you want, and (b) a bunch of tech and philosophy heavy-hitters signed a letter asking for AI research to be paused for 6 months due to the “profound risk to society and humanity.”1
It’s rare for a hot job market to hit the scene out of nowhere in general, but it’s extra rare for that job market to be like “we kinda need… creative writers and interviewers???”So I was kind of excited, but also it makes total sense that the second I finally might catch a break financial success-wise, Steve Wozniak and Elon Musk put aside their differences to tell the world that it will literally end the world.
But while it’s all being debated, I had to try it, and that brings me to question #2:
Isn’t that cheating?
I will answer that by taking you through a brief sampling of the things I’ve used it for and whether or not I think each individual one is cheating.
Gritty Succession. I asked it to “Write a scene of an episode of Succession where Kendall tries to convince Shiv and Roman that they need to partner with the Philadelphia Flyers' Gritty in order to take down their dad.”
Here is an excerpt:
I deem this not cheating. It’s crazy, it’s clearly a stunt, it’s the thing we were all doing with Dall-E where we were like “paint a picture of Binya Binya from Gullah Gullah Island kissing Tim Russert on Meet the Press” and it would just do it.
Plus I’m being very open about how this scene was formulated. But if I used GPT-4 to its fullest extent to write a stunt spec where Gritty teams up with The Succession Kids and they all have an awkwardly silent PJ flight to Richard Yuengling’s Nantucket estate for some party where they need to convince one woman to convince another guy to vote yes on something, and the whole flight Gritty is doing googly eyes at Shiv while getting absolutely plastered on Calvados, that would be more like me coaching or guiding a writer to do my bidding rather than me writing it.
Writing the Sternal Journal. In a classic Sternal Journal Tonal 180, let’s leave behind Gritty takes Manhattan and delve into the serious existential crisis I and all writers are about to “enjoy.” I have an idea for a Sternal Journal where I rank the top podcast theme songs. It is inspired by my love for the tongue-in-cheek, self-deprecating theme song of Joel Stein’s Story of the Week and the many goes-way-too-hard-for-what-it-is themes such as Serial Productions’ The Trojan Horse Affair.
So I asked it to “Write a Substack post in the style of "The Sternal Journal" ranking the top 5 podcast theme songs.”
Here’s the beginning of what it wrote:
To paraphrase another far-more-respected journalist I know and love whose soul I fed into GPT-4, what it blurted out here (with no help from me other than the prompt) is generic, but “terrifyingly serviceable.” It did get that it was supposed to call you all Sternal Journalists, but it frankly was a bit too normal. Also, it’s picks—Radiolab, Reply All, 99PI, Serial, and StartUp—are not close to what I would have picked (I don’t think! But now I’ll never know for sure because I’ve got this version in my head).
If I were to take this, slot out the picks I don’t like and in the ones I do, then add some jokes, and edit it for tone, I think it would be cheating. But if I said, “This story was written with the assistance of GPT-4,” maybe it wouldn’t be? I don’t really know.
Punching Up a Joke. I did it. I asked GPT-4 to make my joke funnier. It is my joke about how one time I was irrationally (and clinically! I have OCD!) scared of being on a flight with a youth soccer team because I felt like them dying in a plane crash would make a very newsworthy headline and therefore it felt inevitable.
I think that, thank God, their version is actually longer and less funny than mine. But “attention universe: perfect tragedy opportunity” is not a terrible alt, and if a friend told me that, I might try it if “Dear God: Smite Here” wasn’t working. But if I tried it? That feels like it would be cheating. This thing has basically been trained on the whole internet. I’m, to borrow a phrase from my wonderful and smart partner Kristen who has had to put up with me talking about this shit nonstop for 48 hours, micro-plagiarizing2.
But aren’t we all micro-plagiarizing when we’re influenced by previous and/or greater people in our fields? And comedy is a world where people give each other joke punch-ups all the time. The argument for this being cheating is that you can’t credit the punch-up to anyone. The argument against is that we rarely give credit when a far-away inspiration cracks a joke or a story or other problem that’s been solved, creative or otherwise. And this is just a new tool that combines every piece of inspiration out there, then regurgitates it based on what you’re looking for. So I deem it… IIII DON’T KNOW!
I don’t think this is specific to stand-up comedy. I think anyone who wants to will be able to find a use for this tool, and it feels like the type of thing that’s just going to be too big to go away. My hope is that we can do it transparently and honestly, and my fear is that I’m being pretty optimistic by even hoping for something so outlandish.
This is about where I should have a conclusion that just wraps it all up, but this feels way too at the beginning of things to wrap anything up. In addition to the examples above, I used GPT-4 to tell me how to cook dinner with the few ingredients I had in the fridge, and tell me how to organize some tasks that I had to get done Friday morning “in the style of a gentle, but practical friend;” I asked it to walk me through an annoying phone call about my health insurance switching providers, and I asked if I was allowed to wear black lapels on a white jacket for the “warm weather black tie” wedding I went to this weekend.
It did it all! It also feels so untethered! I’m really excited about the way its been a tool that is good at organizing my thoughts, but I’m scared of how it could be a tool of telling me what I think. I guess that’s it. I think that’s everything I feel about this crazy thing right now.
For better or for worse, I highly recommend it. Other things I recommend?
Recommendations!
Succession. Television Show. If you like this show already, you know how exciting it is to be enjoying the final season. If you don’t know this show yet, I would argue that it’s an even better binge than it is a week-by-week so I am jealous and get to it!
Warm Weather Black Tie. Dress Code. Anyone who know knows me knows I am not one for dress codes—I described my style to someone today as “Playplace Chic”—but I went to a wedding this weekend that was “warm weather black tie” which mostly just means that guys get to wear white jackets. And I gotta say, if you’re gonna make people dress up, let ‘em reealllllly dress up.
Mockingbird Valley. Song. DJ Drama has a new album out that I haven’t listened to all of yet, but I have obviously listened to the Jack Harlow track. With the album having features from Benny the Butcher, Tyler the Creator, Rick Ross, CyHi, Jeezy, and oodles of other greats, I’m sure it’s not the best, but it is the most Jack Harlow song and that’s why I love it. It’s a nice victory lap/feeling oneself jam.
CALL ME IF YOU GET LOST: The Estate Sale. Album. Speaking of feeling oneself, the deluxe version of Tyler’s CMIYGL is out and it has eight new songs. I have only listened to the full deluxe version once and I listened to the original probably fifteen times, so I’m nowhere having anything special to say about it. Nor do I need to! I think Tyler’s one of the greatest artists of our lifetime so just give it a listen.
That’s all this week! Much love!
Julian
P.S. I think it’s clear that GPT-4 aided me in aspects of this SternJourn, but I think that should be disclosed more formally, which I did here. Mostly, I used it to organize the tasks of writing the Sternal Journal, but it also suggested an outline once I told it what I wanted to write about and gave me a few pieces of feedback which ultimately probably resulted in 2 new sentences and 3 sentences that were very slightly restructured for clarity based on its suggestions. Hopefully, in the near future, there will be better best practices for disclosing this.
P.S. I spend anywhere between two and twelve hours a week on the Sternal Journal. If you enjoy receiving it (and are RICH) consider becoming a paying subscriber. For just a few bucks a month, you can provide me with a bit more time to come up with fun topics, poems, and interviews; and you with probably fewer typos.
It is very worth pointing out that other heavy-hitters think a pause would be pointless or even dangerous, and even others think that actually more than six months is needed.
It is, once again, very worth pointing out that there are a multitude of ethical questions around AI I didn’t cover. When I pushed GPT-4 for feedback on my essay, the possibility to delve into more of the ethics around AI was one area. When I asked what aspects I didn’t discuss, it said the below. And frankly, I think they put it fairly well plus its late, so I’m just gonna share what they said: