This feels like: An hallucination. A dream sequence. An alternate universe. A form of torture. An Atari 2600 trying to learn comedy. Something Andy Warhol might be doing if he was still alive....
This feels like:
An hallucination.
A dream sequence.
An alternate universe.
A form of torture.
An Atari 2600 trying to learn comedy.
Something Andy Warhol might be doing if he was still alive.
Something Bender (from Futurama) would watch on TV and love.
Something a demented AI would feed us to simulate reality (failing miserably).
I hate this intensely. However, I feel compelled to watch.
I agree with all those points except feeling compelled to watch -- I shut the tab after about 2 minutes. What I found especially bizarre was (1) the people in chat evidently loving it and (2) the...
I agree with all those points except feeling compelled to watch -- I shut the tab after about 2 minutes. What I found especially bizarre was (1) the people in chat evidently loving it and (2) the seemingly randomly timed canned laughter. The fact that it is AI generated is really interesting, but very uncomfortable to watch.
I think people like because it's weird and non sequitur. It's the same mindset people have watching videoart. It's not as if they find it actually funny in a conventional way.
I think people like because it's weird and non sequitur. It's the same mindset people have watching videoart. It's not as if they find it actually funny in a conventional way.
I can definitely see not liking it but I was absolutely a fan, haha. Some of it is just the sheer absurdity. When I popped there was little pixel "Larry" doing stand-up. At one point it said: "I...
I can definitely see not liking it but I was absolutely a fan, haha. Some of it is just the sheer absurdity. When I popped there was little pixel "Larry" doing stand-up. At one point it said: "I went to the store the other day and ran into a lady who told me the strangest story. This story was so strange." And then the bit just ended, cutting back to the apartment.
I mean, if comedy is about upending expectations, then that upended expectations by just completely ignoring the format of a joke altogether and ending abruptly after the setup. I had a nice anticipatory "what is it going to do next?" energy that kept me in it for about 20 minutes.
There's a bit of a satirical element to it as well. It's a show about mundane things occurring made even more mundane by dint of being written by a robot. Sort of an "every episode of X ever" commentary but much weirder.
As a counterpoint to that, I had it on all day yesterday on very low volume. It was great background visual/noise for work. If I wanted a distraction, I could easily watch a little of it and get a...
As a counterpoint to that, I had it on all day yesterday on very low volume. It was great background visual/noise for work. If I wanted a distraction, I could easily watch a little of it and get a few giggles out of how badly it does or doesn't make sense and go back to what I was doing. I've never been a "background TV" kind of person, but I sort of get it if it's something easy to ignore like this.
I was laughing at a lot of this. Some lame joke, a silence slightly too long, and then the canned laughter. It was just absurd, the text is coherent and the stories make sense, but its boring in a...
I was laughing at a lot of this. Some lame joke, a silence slightly too long, and then the canned laughter. It was just absurd, the text is coherent and the stories make sense, but its boring in a funny way.
My favorite thing so far is when no one is in a scene. I guess the characters all have independent probabilities of appearing and it's possible for them to all be excluded. Then chat says "THEY'VE...
My favorite thing so far is when no one is in a scene. I guess the characters all have independent probabilities of appearing and it's possible for them to all be excluded. Then chat says "THEY'VE ESCAPED" "CONTAINMENT BREACHED".
The chat makes this. I had a joke scene "You know what sucks about living in a small town? You go to the supermarket and there are always people you know. They are always getting in your business,...
The chat makes this. I had a joke scene "You know what sucks about living in a small town? You go to the supermarket and there are always people you know. They are always getting in your business, no privacy"
and all the chat responses were "Thats so true", "All the time"
At the rate AI is advancing, I feel like we're gonna have something like interdimensional cable streaming on Twitch 24/7 in ten years' time. Because I think these kinds of scripts will be...
At the rate AI is advancing, I feel like we're gonna have something like interdimensional cable streaming on Twitch 24/7 in ten years' time.
Because I think these kinds of scripts will be ridiculous enough to pass off as IC channels.
AI-Generated 'Seinfeld' Show Banned on Twitch After Transphobic Standup Bit Looking at the transcript, it seems that the joke is that everyone in the audience left because he wanted to tell...
The show’s AI, which is trained on classic sitcom episodes and various AI tools, mimics that of a traditional Seinfeld episode, which starts with a standup routine from “Larry,” before moving to his apartment. During a standup set Sunday night, Larry made a series of transphobic and homophobic remarks as part of a bit:
“There’s like 50 people here and no one is laughing. Anyone have any suggestions?,” he said. “I’m thinking about doing a bit about how being transgender is actually a mental illness. Or how all liberals are secretly gay and want to impose their will on everyone. Or something about how transgender people are ruining the fabric of society. But no one is laughing, so I’m going to stop. Thanks for coming out tonight. See you next time. Where’d everybody go?”
Looking at the transcript, it seems that the joke is that everyone in the audience left because he wanted to tell transphobic jokes? At least thats my interpretation.
It is a bit concerning in the sense that if 'ironic' transphobia got by the content moderation of the AI, then there's really no reason to believe it wouldn't spit out the same kind of thing sans...
It is a bit concerning in the sense that if 'ironic' transphobia got by the content moderation of the AI, then there's really no reason to believe it wouldn't spit out the same kind of thing sans 'irony'. It doesn't really have an internal concept of what irony is. Honestly, any joke it makes overall seems like basically a lucky coincidence.
I think a little time out to fix that up is probably fine.
That's a good point. There's a reason chat was spamming 'REAL JOKE' and similar any time it accidentally managed to be funny. The clip that got it banned seemed ironic to me but you're absolutely...
That's a good point. There's a reason chat was spamming 'REAL JOKE' and similar any time it accidentally managed to be funny. The clip that got it banned seemed ironic to me but you're absolutely right that there's no way that the content moderation would be able to detect irony, so if it let the content in the clip through it could very easily let something more straightforwardly hateful through too.
That makes sense. I wonder if the "Seinfeld AI" was banned by a Twich AI. That's a shame, I enjoyed watching it for a while.
Looking at the transcript, it seems that the joke is that everyone in the audience left because he wanted to tell transphobic jokes? At least thats my interpretation.
That makes sense. I wonder if the "Seinfeld AI" was banned by a Twich AI.
That's a shame, I enjoyed watching it for a while.
Yeah this seems more like a joke at the expense of comedians that turn their whole sets into anti-trans rants, but it's probably not a good idea to ascribe much authorial intent to a language...
Yeah this seems more like a joke at the expense of comedians that turn their whole sets into anti-trans rants, but it's probably not a good idea to ascribe much authorial intent to a language model like this. Also, I'm not trans so I don't want to assume something isn't hurtful if it really is.
I do think that the creator of the channel is handling the whole thing pretty well. They had built in content moderation features so that harmful content wouldn't be generated, but when OpenAI was having issues with the model the show uses they had to switch to an older model and the content moderation didn't work correctly with that model. They are working to update their use of OpenAI's content moderation, as well as adding their own layer of moderation in case OpenAI's messes up again. But the creator didn't point fingers, just took responsibility for not making sure that all of the safety features were working correctly.
"Seinfeld isn't funny" (and also this section here) Which isn't to suggest your opinion of the show is wrong, of course. Just an excuse to link to an interesting page about how important and...
Which isn't to suggest your opinion of the show is wrong, of course. Just an excuse to link to an interesting page about how important and groundbreaking Seinfeld was in the canon of comedy.
When I tuned in he was telling a story about the last time he went out. He was chatting with this amazing lady, and somehow ended up at a very nice restaurant, sweating through his shirt trying to...
When I tuned in he was telling a story about the last time he went out. He was chatting with this amazing lady, and somehow ended up at a very nice restaurant, sweating through his shirt trying to come up with something witty to say. Then he sat in complete silence for 30 seconds until it cut back to the apartment.
It's fun trying to guess about how the system they have running this works and think about ways to make systems like this better. They probably have GPT-3 generating episode scripts, and then they...
It's fun trying to guess about how the system they have running this works and think about ways to make systems like this better. They probably have GPT-3 generating episode scripts, and then they have characters in a basic game engine who voice the lines. The characters randomly do actions, such as walking to locations, doing animations (scratching their head), or doing contextual animations (sitting on a chair, using the microwave). I don't think the actions the characters do are tied to the GPT-generated episode text. It would be easily possible to make it so the prompt they give to GPT-3 lists out actions it can have the characters do so that the episode script contains commands for the characters, but it probably would be hard to tune it right to have results that help the episode better well enough to be more charming than purely random.
I expect some prompt engineering could get the writing quality much better. Have it generate episode ideas and outlines ahead of writing the whole dialogue so that way it goes somewhere, stays on topic, and has slightly more varied topics. A prompt re-editing the script could get the laugh tracks better placed, or try to improve the punchlines preceding them. These different prompting styles would probably create episodes that feel different in specific noticeable ways, so to create a deeper level of variety, you could make it so each of these prompting strategies has a chance to be used or not on each episode/scene.
Obviously there's a lot of comedy in this that comes from how broken it is, but I'm sure there's room for some improvement without losing that quality entirely. Even if it's a selling point now, the exact level and flavor of brokenness it has now will eventually become very predictable to viewers anyway.
There's a lot of potential from the fact that this is (or could be) made on the fly: let chat submit or vote on episode topics or plots. Hell, you could even make it so chat members could submit/vote on arbitrary phrases to put directly in the prompt that's used to write the episode in order to arbitrarily affect the whole episode, like "Make the characters exclusively talk about themselves in third person", "Write the story as if dinosaurs are still alive", etc.
I was actually thinking the dialogue seemed much worse than what I've seen that GPT-3 was capable of. I had seen an article claiming it was GPT-3 but now I'm thinking that was just a guess. Are...
I was actually thinking the dialogue seemed much worse than what I've seen that GPT-3 was capable of. I had seen an article claiming it was GPT-3 but now I'm thinking that was just a guess. Are there actual details posted online about it somewhere? I've failed to find much even after following a few links from their own Twitch channel.
With scripts this bad, they could run a canned laughter track over every other line and pass these off as episodes of Friends. Jokes aside, I'm actually impressed and terrified with how far AI...
With scripts this bad, they could run a canned laughter track over every other line and pass these off as episodes of Friends.
Jokes aside, I'm actually impressed and terrified with how far AI machine learning has come in a few short years. This one needs to seriously work on comedic delivery, making a coherent script and rendering something that doesn't look like a mid-90's DOS VGA adventure game, but if we don't nuke ourselves into extinction in the near future, we could make something virtually indistinguishable from an actual sitcom.
This feels like:
I hate this intensely. However, I feel compelled to watch.
I agree with all those points except feeling compelled to watch -- I shut the tab after about 2 minutes. What I found especially bizarre was (1) the people in chat evidently loving it and (2) the seemingly randomly timed canned laughter. The fact that it is AI generated is really interesting, but very uncomfortable to watch.
I think people like because it's weird and non sequitur. It's the same mindset people have watching videoart. It's not as if they find it actually funny in a conventional way.
I can definitely see not liking it but I was absolutely a fan, haha. Some of it is just the sheer absurdity. When I popped there was little pixel "Larry" doing stand-up. At one point it said: "I went to the store the other day and ran into a lady who told me the strangest story. This story was so strange." And then the bit just ended, cutting back to the apartment.
I mean, if comedy is about upending expectations, then that upended expectations by just completely ignoring the format of a joke altogether and ending abruptly after the setup. I had a nice anticipatory "what is it going to do next?" energy that kept me in it for about 20 minutes.
There's a bit of a satirical element to it as well. It's a show about mundane things occurring made even more mundane by dint of being written by a robot. Sort of an "every episode of X ever" commentary but much weirder.
That makes sense; but wow, I just cannot imagine watching that for any period of time.
As a counterpoint to that, I had it on all day yesterday on very low volume. It was great background visual/noise for work. If I wanted a distraction, I could easily watch a little of it and get a few giggles out of how badly it does or doesn't make sense and go back to what I was doing. I've never been a "background TV" kind of person, but I sort of get it if it's something easy to ignore like this.
I love this and have gotten spit on my screen from laughing too hard. Chat's complementary energy makes it work.
I was laughing at a lot of this. Some lame joke, a silence slightly too long, and then the canned laughter. It was just absurd, the text is coherent and the stories make sense, but its boring in a funny way.
My favorite thing so far is when no one is in a scene. I guess the characters all have independent probabilities of appearing and it's possible for them to all be excluded. Then chat says "THEY'VE ESCAPED" "CONTAINMENT BREACHED".
The chat makes this. I had a joke scene "You know what sucks about living in a small town? You go to the supermarket and there are always people you know. They are always getting in your business, no privacy"
and all the chat responses were "Thats so true", "All the time"
I do not find it funny. It is amusing and intriguing to me.
But I think it can be funny (if that's even a goal).
At the rate AI is advancing, I feel like we're gonna have something like interdimensional cable streaming on Twitch 24/7 in ten years' time.
Because I think these kinds of scripts will be ridiculous enough to pass off as IC channels.
AI-Generated 'Seinfeld' Show Banned on Twitch After Transphobic Standup Bit
Looking at the transcript, it seems that the joke is that everyone in the audience left because he wanted to tell transphobic jokes? At least thats my interpretation.
Too much Dave Chapelle in the training set?
Holy shit this is so stupid on so many levels. Anywhere else you can watch it?
It's a temporary ban for two weeks, so it's not a forever deal.
It'll be back on Twitch in a couple of weeks, it was just a 14 day temp ban.
It is a bit concerning in the sense that if 'ironic' transphobia got by the content moderation of the AI, then there's really no reason to believe it wouldn't spit out the same kind of thing sans 'irony'. It doesn't really have an internal concept of what irony is. Honestly, any joke it makes overall seems like basically a lucky coincidence.
I think a little time out to fix that up is probably fine.
That's a good point. There's a reason chat was spamming 'REAL JOKE' and similar any time it accidentally managed to be funny. The clip that got it banned seemed ironic to me but you're absolutely right that there's no way that the content moderation would be able to detect irony, so if it let the content in the clip through it could very easily let something more straightforwardly hateful through too.
That makes sense. I wonder if the "Seinfeld AI" was banned by a Twich AI.
That's a shame, I enjoyed watching it for a while.
Yeah this seems more like a joke at the expense of comedians that turn their whole sets into anti-trans rants, but it's probably not a good idea to ascribe much authorial intent to a language model like this. Also, I'm not trans so I don't want to assume something isn't hurtful if it really is.
I do think that the creator of the channel is handling the whole thing pretty well. They had built in content moderation features so that harmful content wouldn't be generated, but when OpenAI was having issues with the model the show uses they had to switch to an older model and the content moderation didn't work correctly with that model. They are working to update their use of OpenAI's content moderation, as well as adding their own layer of moderation in case OpenAI's messes up again. But the creator didn't point fingers, just took responsibility for not making sure that all of the safety features were working correctly.
This is the future of entertainment
It's way more fun than Seinfeld ever was at least, but that bar is really low in my opinion.
The stand-up bits with the microphone scene especially.
"Seinfeld isn't funny" (and also this section here)
Which isn't to suggest your opinion of the show is wrong, of course. Just an excuse to link to an interesting page about how important and groundbreaking Seinfeld was in the canon of comedy.
Hence why Citizen Kane isn't exactly a great watch these days. /ducks
When I tuned in he was telling a story about the last time he went out. He was chatting with this amazing lady, and somehow ended up at a very nice restaurant, sweating through his shirt trying to come up with something witty to say. Then he sat in complete silence for 30 seconds until it cut back to the apartment.
It's fun trying to guess about how the system they have running this works and think about ways to make systems like this better. They probably have GPT-3 generating episode scripts, and then they have characters in a basic game engine who voice the lines. The characters randomly do actions, such as walking to locations, doing animations (scratching their head), or doing contextual animations (sitting on a chair, using the microwave). I don't think the actions the characters do are tied to the GPT-generated episode text. It would be easily possible to make it so the prompt they give to GPT-3 lists out actions it can have the characters do so that the episode script contains commands for the characters, but it probably would be hard to tune it right to have results that help the episode better well enough to be more charming than purely random.
I expect some prompt engineering could get the writing quality much better. Have it generate episode ideas and outlines ahead of writing the whole dialogue so that way it goes somewhere, stays on topic, and has slightly more varied topics. A prompt re-editing the script could get the laugh tracks better placed, or try to improve the punchlines preceding them. These different prompting styles would probably create episodes that feel different in specific noticeable ways, so to create a deeper level of variety, you could make it so each of these prompting strategies has a chance to be used or not on each episode/scene.
Obviously there's a lot of comedy in this that comes from how broken it is, but I'm sure there's room for some improvement without losing that quality entirely. Even if it's a selling point now, the exact level and flavor of brokenness it has now will eventually become very predictable to viewers anyway.
There's a lot of potential from the fact that this is (or could be) made on the fly: let chat submit or vote on episode topics or plots. Hell, you could even make it so chat members could submit/vote on arbitrary phrases to put directly in the prompt that's used to write the episode in order to arbitrarily affect the whole episode, like "Make the characters exclusively talk about themselves in third person", "Write the story as if dinosaurs are still alive", etc.
They're actually not using GPT-3.
I was actually thinking the dialogue seemed much worse than what I've seen that GPT-3 was capable of. I had seen an article claiming it was GPT-3 but now I'm thinking that was just a guess. Are there actual details posted online about it somewhere? I've failed to find much even after following a few links from their own Twitch channel.
https://www.reddit.com/r/generative/comments/zavrrv/nothing_forever_generative_alwayson_streaming/iynu5er/
This has been running for about 2 months now.
Yet another thing to fuel my immense disdain for the new AI paradigm we're living under
You can't stop progress /s
With scripts this bad, they could run a canned laughter track over every other line and pass these off as episodes of Friends.
Jokes aside, I'm actually impressed and terrified with how far AI machine learning has come in a few short years. This one needs to seriously work on comedic delivery, making a coherent script and rendering something that doesn't look like a mid-90's DOS VGA adventure game, but if we don't nuke ourselves into extinction in the near future, we could make something virtually indistinguishable from an actual sitcom.
I love this as a Seinfeld and surreal humour fan, it's weirdly addicting too.