Generative AI is out of control: Nothing, Forever is a Seinfeld spoof about nothing... forever
It's... actually kind of funny, though*
Updated When plague winds howl across the surface of cadaver world Earth, humanity long dead by its own hand, imperial archaeologists picking through the remnants will excavate a bunker. Inside they will find a primitive computer. The computer broadcasts AI-generated spoof Seinfeld episodes for eternity.
This is the haunting promise of the Twitch channel watchmeforever officially launched last week by Mismatch Media, which claims to be about "experimental forms of television shows, video games, and more, through generative... and other machine learning technologies."
Twitch is a good place for "experimental" TV because that's exactly what it is. Viewers can interact with livestreamed video in real time, though the vast bulk of broadcasters use it to show off their skill (or lack thereof) at video games. There are, however, some "visionary" streamers out there and, while we hate to admit it, watchmeforever may be one of them.
The "show" is called Nothing, Forever and stars the ensemble cast of Larry Feinberg, Yvonne Torres, Fred Kastopolous, and Zoltan Kakler.
Sound familiar? While fans of Seinfeld and Curb Your Enthusiasm may be among the first to admit that the Larry David sitcoms aren't really about anything at all, at least they are hilarious by virtue of the characters' mundane and inept interactions.
Nothing, Forever has no such saving grace. Why? "Much of our generative content is provided through OpenAI's GPT-3," says Mismatch Media. In fact, outlines for Seinfeld episodes have become a favorite subject among those tinkering with the more consumer-facing ChatGPT so it's easy to see where the genesis of this bizarre project came from. Mismatch says of the show:
Nothing, Forever is a show about nothing, that happens forever. Kinda like popular sitcoms of the past, except that it never stops.
Nothing, Forever is always-on, runs 365 days of the year, and delivers new content every minute.
Everything you see, hear, or experience (with the exception of the artwork and laugh track) is always brand new content, generated via machine learning and AI algorithms.
So what's it like? Look at it this way – sci-fi of decades past imagined AI either elevating humanity or destroying it. Nothing, Forever proves that it is time to leave those conceptions at the door for another 30 years or so because it's toe-curlingly bad.
- Tech CEO nixes AI lawyer stunt after being threatened with jail time
- So you want to replace workers with AI? Watch out for retraining fees, they're a killer
- AI cannot be credited as authors in papers, top academic journals rule
- ChatGPT talks its way through Wharton MBA, medical exams
But, and it's a big but, we need to remember that it is produced by GPT-3 and isn't trying to be good. Combined with graphics reminiscent of Dire Straits' "Money for Nothing" video and Twitch's permanently deranged chat, Nothing, Forever is rife with forehead-slapping moments of infuriating faux pas. Here's a summary of a particularly memorable skit, the likes of which we will never see again:
Yvonne announces to those present in Larry's apartment that she's finally put her new blinds up. Larry asks if she chose a "good color" for them and she tells him she picked orange. When pressed for her reasoning, she says it makes the place "bright." Larry berates her for not being able to think of a better word, but Fred comes to her defense, insisting that "orange is a modern color, Larry," and, in a strangely prescient moment of clarity for GPT-3, says "orange is the new black." They argue briefly about which way round that phrase should be.
There's another where Yvonne tells Larry that they should "do something spontaneous" like "swimming at the beach." Larry enthusiastically agrees, saying his swimsuit is already in the car. The funny thing is that you kind of get excited to see where the scene is headed, but there's never any resolution – it just cuts to the outside of the apartment with an AI-mangled attempt at Seinfeld's iconic slap bass jingles.
The next scene returns to the interior with the characters on a completely unrelated tangent, probably about a "new restaurant" with some sort of gimmicky menu – a recurring theme – and the characters debating whether they should go.
These are bookended with the classic "Larry in front of the red curtain" standup sections, which are equally insane. In one, Larry quips: "Why did the chicken cross the playground? To get to the other slide! HA HA HA." He follows this up with: "What do you call a bear with no teeth?"
But we never get the punchline (we presumes it's "a gummy bear"). It's back to the apartment ready for another weird interaction, like Kakler saying his new method of tying shoelaces will leave Yvonne "in stitches" but he goes on to explain to her that she should always do them in a double knot. To add to the madness, Larry often inserts [PAUSE FOR AUDIENCE REACTION] into his standup routine as if he's waiting for the stuttering laugh track to come in, and characters regularly clip through the furniture and each other. For example, we witnessed Fred abruptly choose to sit inside Yvonne to much mirth in chat.
The Register asked Mismatch Media if it could tell us more about the endeavor, like how the model was trained, and we'll update the piece if we get a response. We guess they're out at the minute because Nothing, Forever is pretty much the ultimate "set it and forget it" software project.
Updated to add:
* The parts we watched, of course.
This article was written before the Twitch channel got shut down on 6 February, allegedly over a transphobic standup bit.
In the project's Discord, the show's staff blamed the event for having to switch AI models, which resulted in "errant behavior."
"We investigated the root cause of the problem," Tinylobsta, a contributor, reportedly wrote on Discord (the channel is no longer accessible). "Earlier tonight we started a failure when using OpenAI's GPT-3 Davinci model, which caused the show to exhibit erratic behavior (you may have seen empty spaces cycling through).
"OpenAI has a less sophisticated model, Curie, which was the predecessor of Davinci. When davinci started failing we switched to Curie to try and keep the show going with no downtime. Switching to Curie caused the inappropriate text to be generated.
"We use OpenAI's content moderation tools, which have previously worked for the Davinci model but have not worked with Curie. We were able to identify the cause of our problem with the Davinci model and will not use Curie as a fallback in the future."
Another contributor added:
"I would like to add that nothing that has been said reflects the opinion of the developers (or anyone else on the staff team)." ®