13
votes
Researchers train and apply an LLM and an image generator to create bespoke South Park episodes
Link information
This data is scraped automatically and may be incorrect.
- Title
- SHOW-1 and Showrunner Agents in Multi-Agent Simulations
- Word count
- 4215 words
This couldn't have worse timing with the ongoing writers strike. Personally, I would still rather watch something hand-crafted by artists. And yet, as soon as it's easy and cheap enough to use, I will certainly watch a rewritten (and extended) final season(s) of Game of Thrones.
While it is an option, I'm not sure of how viable it will all be. I remember digital cameras/photo software would turn every average citizen into an award winning photog. Or when word processors got spell and grammar check meant that every person would turn into Hemmingway. I'm sure they said that moving pictures would make theater obsolete since you could do 1 recording of Nutcracker or whatever and see it multiple times (same with the phonograph)..
I think AI will help great artists stay great as they learn the tools. It might elevate good artists into great ones. I'm sure there will be a few new "great" artists created by using AI. However, For every 1 "great" I'd wager there are hundreds of thousands or millions average people who will dabble in AI anything more than maybe "Mickey Mouse does politician X" prompts or other useless and unoriginal concepts.
I may very well be wrong and approaching it from the wrong side, but I think "Garbage In, Garbage Out" is still very applicable.
It’s not a very good episode of South Park. The jokes don’t land, and they aren’t nearly as clever as actual South Park jokes. It does a good enough job mimicking the flow of a South Park episode, but it’s all empty calories. I don’t really see how the tech that wrote this episode could possibly write a better final season of GoT or an episode of anything that would actually be fun or engaging to watch.
I’m certainly not judging on the quality of what is being presented this week. That’s besides the point. This is only a proof of concept.
How long has GPT4 and Stable Diffusion been around? Compare what you see today to what we had 6 months ago.
Remember, this is the worst quality example.
And it's improving exponentially.
And these AI have 70 billion+ parameters. It's getting into the order of magnitude of human neuron count. Although we will probably have to more like 10x it because neurons do a LOT more than parameters.
'Mett Porker' is gold though.
Also, Stan and Kyle complaining about AI-written shows and characters while being AI-written characters themselves seems exactly like the kind of sarcastic fourth-wall-breaking humor that South Park would do.
It was definitely janky and repetitive at times, but for an early proof of concept I think it did a passable job of mimicking the South Park flow and humor.
Only tangentially related, but my prediction is that "bespoke" content will be everywhere, starting with programs. Forget software as a service where you have someone's pre-made code on the cloud, prepare for bespoke-software-as-a-service where you'll log into the portal and you'll just be ordering up whatever program you need, coded and created on the fly by AI.
"AI, give me a home budgeting tool, give me a social media crawler that finds all X".
Bespoke movies will need A LOT of resources. After all you have to render everything. Bespoke code is just code. If the AI is good at creating it, it can do so with minimal resources.
Edit: and that's probably just a step towards a full AI-assistant where you just say what you need and the AI creates the tools to find, compile and present the information to you, instead of running a simple web search like say Alexa or whatever.