Explicitly flagging this llm as a tool for (potentially) writing fiction is a nice way to work on domain-specific text generation while avoiding many of the critiques of Galactica.
Yes, this is a companion to https://arxiv.org/abs/2209.14958: "We share many reflections in our published pre-print (Mirowski, Mathewson et al (2022) “Co-Writing Screenplays and Theatre Scripts with Language Models: An Evaluation by Industry Professionals.”)"
i m sure the model is great, but industrially produced scripts sounds extremely offputting. Art is communication and nobody wants to communicate with a bot. Maybe this would work as an endless roleplaying game or something but theater and film scripts? sounds awful
> The stuff ChatGPT comes up with sounds funnier and more entertaining than 90% of the current Netflix catalog.
I just made it write a skit between r/wallstreetbets and Bill Burr during the GME saga... it was not as funny as when I then asked it to write a skit where Bill starts shilling for it.
It might have been something the film adaptation of Wolf of Wall street left on the cutting room floor, even down to the facial descriptors. Try it for yourself, its pretty silly, but so was Wolf of Wall street, and I read the book after the movie because I was told it had more gonzo-like story telling.
Tell me a story about ___. Every 2 paragraphs show a very descriptive Dalle-2 prompt for the image which fits into the story.
The result was pretty interesting I thought. ChatGPT did a good job here. I bet it would have been even better with 2022's additional prompt samples found on the internet.
> The result was pretty interesting I thought. ChatGPT did a good job here. I bet it would have been even better with 2022's additional prompt samples found on the internet.
I never got access to DALLE, so I will never know for sure, but this is disrupting visual learning/story telling in real time, so this coupled with a decent sized social media followers and pateron that tech is finally delivering on what the Internet was promising since the days of Geo-sites.
I think writing children's books for example has just gotten easier than ever for a very accessible price (~$15).
Imagine providing a complicated visual shape as input and the LLM just generates a story which fits the shape. The shape could be subdivided into patches. We could specify the textures of these patches which could represent emotions, style, wit, pace, etc all blending nicely. So many possibilities. Writers never have to worry about continuity issues or tracking details in their head. In future, book writing would be similar to game development.
why would someone read the book though instead of playing the game? At least in the game there is a human (the player). I think reading needs to be a meaningful action
If the story is identical in game format or book format, why would someone prefer reading a book vs playing the game. List all possible reasons that come to your mind.
There are a few reasons why someone might prefer reading a book over playing a game, even if the story is the same in both formats. Some possible reasons include:
Reading a book allows a person to use their imagination to visualize the story and create their own mental images of the characters and settings, whereas a game may have pre-determined visuals that the player has no control over.
Reading a book can be a more passive activity, allowing the reader to relax and immerse themselves in the story without having to actively engage with the narrative by making decisions or solving puzzles.
Reading a book can be a more personal experience, as the reader can read at their own pace and take their time to reflect on the story and its themes.
Reading a book may be a more affordable option than purchasing and playing a game, especially if the reader already owns the book or has access to it through a library or other resource.
Some people simply prefer reading as a means of entertainment and may find it more enjoyable than playing a game.
Reading a book can be a more portable and convenient option, as it doesn't require access to a gaming device or a stable internet connection.
Is "nobody wants to communicate with a bot" off-putting to you because bot content isn't quality enough yet or because of the fact it's from a bot? I'm firmly in the former camp but curious what you meant there. For me the attractiveness of art isn't that the feelings I get came from a person and that's why I liked the art as much as the feelings I get were interesting/fun/unique/etc.
It’s the difference between a mass produced product and a hand made artisan one. Sure you can make the former look like the latter, but eventually you arrive at a boring world that has no uniqueness if you eliminate all human creativity. Mass production is a miracle that has enabled billions of people to enjoy everything from clothing to mobile phones, but humans crave variety and novelty almost on a biological level and without the artisans we’d run out of both quite quickly.
What makes mass produced bland is not that it lacks human creativity it's it's the exact same model and color iPhone * 10,000,000 because it's cheaper. You can just as easily plug in something else with creativity and not pay the lowest possible price as you can plug in human creativity and not pay the lowest possible price. Current ML models aren't great at coming up with their own variations on things and if you want to call that "human creativity" then sure but I'm firmly in the camp that type of creativity isn't something forever bound to being human for the rest of time.
I had this same thought. It makes sense that they would leverage gpt3, given how expensive it was to train. I wonder if they see themselves as competitors?
Except, you know, a child has feelings, and a deep learning model does not. The whole reason it would be wrong to tell your kid that is because your kid is a person with feelings.
> Perspective is a free API that helps you host better conversations online. The API uses machine learning models to score the perceived impact a comment might have on a conversation.
Each is connected to a prompt and uses content from the previous text input to complete a the prompt for the next input. Inputs are editable at each stage. AKA Prompt chaining.
The "Settings" prompt returned nothing. And the "dialogue" prompt hung up, still saying "Generating..." for five minutes now. And there isn't any other option but to refresh the page.
Still going to check out the attached 100-page research paper, but not as urgently since testing out the app.
I'm starting to lose track of the need for all this "assisted stuff".
People have written the most incredible scripts without "the help of AI". If someone isn't a great writer, do they need to be augmented into great writers so every script is great using "AI(tm)"?
Is the plan that we feed AI all the greatest scripts ever written so writing becomes a toy, we destroy writing as a job and then just have eveyone using "aiwriter.com" to generate scripts like it's a toy?
I think like many things, being a good writer is more than just producing a good scripts...
I think this, and AI doomerism in general, greatly underestimates humans and biology in general. Humans and biological life didn’t get where they are because competition was non-existent.
We don't destroy writing as a job. We improve the baseline. So someone who cannot write at all can produce something slightly decent rather than nothing at all. Also this can be useful for inspiration, too.
It's unlikely that the AI will create the greatest scripts for a very long time. Even if it does, it will be like it's "just one more script writer".
IKEA didn't kill hand made furnitures. But it helps some people get some nice stuff.
Hard pass. Netflix is already practically generating micro-targeted blocks of content on an assembly line and now bots are going to write the scripts too? No thank you.
There's a game script I've tried writing a few times but never gotten far on. When I found this paper a few weeks ago my first idea was to see if it could help. I followed the prompt instructions using davinci-002, spending about $0.36 in total, and was a little disappointed in the results. Granted, to save credits, I deleted my previous inputs each time, so when it got to generating dialog for each of the scenes, "previous beat:" wasn't quite enough context to prevent lines from repeating themselves. I suppose these professional writers have better vision than I do if they think they can "take it to Netflix with a few edits". Maybe I'll try it again with ChatGPT.
The best AI written stories I've read have been in the character.ai chat rooms. The dialogue is amazingly natural. Just this morning I created a character of a disgraced syrup mascot out of two sentences from old advertisements. Auntie talked about noodling for catfish with a Gawr Gura bot until I was crying from laughter. Last week I made Gura talk with a "human" Mark Zuckerberg until I started hiccuping from laughter. Maybe it's my just my sense of humor.
Btw, did anyone else try to reproduce the results themselves prior to this tool? The PDF formatting was a little wonky so I tried seeing if anyone posted the plaintext prompts anywhere, only to find zero results on the internet. I'm thankful for the work being done here and will be following this closely. I wish I could have seen the plays live!
48 comments
[ 2.5 ms ] story [ 123 ms ] threadThe number of ChatGPT submissions in the past week begs to differ.
I just made it write a skit between r/wallstreetbets and Bill Burr during the GME saga... it was not as funny as when I then asked it to write a skit where Bill starts shilling for it.
It might have been something the film adaptation of Wolf of Wall street left on the cutting room floor, even down to the facial descriptors. Try it for yourself, its pretty silly, but so was Wolf of Wall street, and I read the book after the movie because I was told it had more gonzo-like story telling.
Tell me a story about ___. Every 2 paragraphs show a very descriptive Dalle-2 prompt for the image which fits into the story.
The result was pretty interesting I thought. ChatGPT did a good job here. I bet it would have been even better with 2022's additional prompt samples found on the internet.
I never got access to DALLE, so I will never know for sure, but this is disrupting visual learning/story telling in real time, so this coupled with a decent sized social media followers and pateron that tech is finally delivering on what the Internet was promising since the days of Geo-sites.
I think writing children's books for example has just gotten easier than ever for a very accessible price (~$15).
Update: It happened [0].
0: https://www.youtube.com/watch?v=_KW1XiLEt_M
There are a few reasons why someone might prefer reading a book over playing a game, even if the story is the same in both formats. Some possible reasons include:
Reading a book allows a person to use their imagination to visualize the story and create their own mental images of the characters and settings, whereas a game may have pre-determined visuals that the player has no control over.
Reading a book can be a more passive activity, allowing the reader to relax and immerse themselves in the story without having to actively engage with the narrative by making decisions or solving puzzles.
Reading a book can be a more personal experience, as the reader can read at their own pace and take their time to reflect on the story and its themes.
Reading a book may be a more affordable option than purchasing and playing a game, especially if the reader already owns the book or has access to it through a library or other resource.
Some people simply prefer reading as a means of entertainment and may find it more enjoyable than playing a game.
Reading a book can be a more portable and convenient option, as it doesn't require access to a gaming device or a stable internet connection.
- answer by ChatGPT.
> Perspective is a free API that helps you host better conversations online. The API uses machine learning models to score the perceived impact a comment might have on a conversation.
Anyone here have experience with it?
It's literally just text inputs.
Each is connected to a prompt and uses content from the previous text input to complete a the prompt for the next input. Inputs are editable at each stage. AKA Prompt chaining.
The "Settings" prompt returned nothing. And the "dialogue" prompt hung up, still saying "Generating..." for five minutes now. And there isn't any other option but to refresh the page.
Still going to check out the attached 100-page research paper, but not as urgently since testing out the app.
Probably not even a "20% project", more like a "2%" project of these deepminders.
Gotta go - need to hack something like this but for branching dialogue trees
Video of DeepMind Dramatron (Chinchilla version):
https://www.youtube.com/watch?v=i2WEEZ03Nu0
People have written the most incredible scripts without "the help of AI". If someone isn't a great writer, do they need to be augmented into great writers so every script is great using "AI(tm)"?
Is the plan that we feed AI all the greatest scripts ever written so writing becomes a toy, we destroy writing as a job and then just have eveyone using "aiwriter.com" to generate scripts like it's a toy?
I think like many things, being a good writer is more than just producing a good scripts...
Makes me wonder where it's all headed.
I mean I don't understand the value proposition of all these tools.
In large part, being good at something is not just about producing text, it's about context, style etc.
Just having a tool which gives you something "pretty good" doesn't mean much on it's own.
It's unlikely that the AI will create the greatest scripts for a very long time. Even if it does, it will be like it's "just one more script writer".
IKEA didn't kill hand made furnitures. But it helps some people get some nice stuff.
The best AI written stories I've read have been in the character.ai chat rooms. The dialogue is amazingly natural. Just this morning I created a character of a disgraced syrup mascot out of two sentences from old advertisements. Auntie talked about noodling for catfish with a Gawr Gura bot until I was crying from laughter. Last week I made Gura talk with a "human" Mark Zuckerberg until I started hiccuping from laughter. Maybe it's my just my sense of humor.
Btw, did anyone else try to reproduce the results themselves prior to this tool? The PDF formatting was a little wonky so I tried seeing if anyone posted the plaintext prompts anywhere, only to find zero results on the internet. I'm thankful for the work being done here and will be following this closely. I wish I could have seen the plays live!