I started with something more “contained” and easier to manage because actual users will go off script every chance they get, and this is basically like playing chess against yourself while reading a book on how to play chess. But, I may have found a kind of working compromise in terms of format and what needs to be captured. Will need a few days to see how it holds up, but right now, this is the basic idea:
Initial PROMPT to get the story started, followed by THOUGHTS that examine them from a gaming perspective, an ACTION, my THOUGHTS, another PROMPT, and.. this is where I was having a tough time because some of the mechanics were not being captured in the THOUGHTS prior. It was only as I wrote the PROMPT that I figured-out certain details or actions that needed to be in play. So when I write a PROMPT that contains these other elements, I write a LOGIC section below them to explain why I “prompted” the way I did.
In crafting the story as you go, the PROMPT is also part of the THOUGHT process! I’m sure anyone giving this a try will be writing and re-writing their prompt as part of the process. Having this extra LOGIC step seems to clean that up, but I don’t think any ML algo will ever keep track of story elements, have ideas on where to take the story next, and then backtrack. Perhaps the “prompt” is some adversarial output from the thoughts, but still internal to process, leading to more thoughts (aka the logic), which leads to the actual output.
Found a rhythm using PLAT (Prompt. Logic. Action. Thought.) but am only averaging 185 words per step. That would be about 18,000 words for 100 steps, or 54,000 words for 300 (which is the very bottom end of book territory). Agree that 100 steps is no story, but waiting to reach 100 steps before checking-in is waiting to long.
Would recommend anyone near the 20 step or 10 pages mark send that in for feedback before going further. I’m going to ignore my own advice because I’d like to complete the first 3 scenes, which is closer to 10% of the full story.
People are concerned about upfront time commitment, while also being focused on 100 step minimum. In another comment I went over how 250 steps works as a better minimum, but to keep all the numbers aligned, perhaps every story should be in 3 acts of 100 steps each (with at least 300 steps being a requirement; handing-in 299 steps would seem sloppy and rude). That would make each “short” story worth $6k, and each act $2k, which is the same 10% of 1,000 steps. Except, handing in the first act should only reserve your $6k payout, not result in getting $2k at a time (desire for finished products and not having to burden anyone with increased tracking/management). There could also be an $18k cap (for 3 short stories of at least 300 steps each) to both limit the number of short stories submitted and let people know there is no “dominating” of the short story space.
With no idea what the arc of the run/story will be, it’s really hard to plan for 3 acts, so maybe not so useful. But did want to leave another comment about scenes. With 4 scenes being about 50 steps, just as a reference, we can look at the number of scenes in a movie to figure each run could be 500 to 750 steps in total length. I just don’t see 1,000 steps as being anything other than an arbitrary dataset requirement. 250-300 steps as a playable run. 500 to 600 steps as a “movie length” representation. And then to double that?
The mental requirement to “film” a Lord of the Rings trilogy while also “filming” the behind the scenes of that filming and also “filming” the real-time documentary required to keep track of everything… while not being clear on how that extra “run time” translates into being better training data.
Is there going to be a “THIS” post, using sample work that you really like and “demanding” all other entries follow that exact format? How will variations in formatting be addressed? Does it need to be?
If you get something that checks all the right boxes, with one exception that leads to a rejection, I think we’d all like to know what that one must-have is.
Using scenes as a marker has some added benefit as I find myself leaving high level comments about some of the next scenes (I had nothing planned beyond the start, but the natural progression leads to speculation about future events or details). This is some of that looking ahead data that this project wanted to capture. Perhaps there should be a FUTURE keyword to wrap these things under? It would basically be a THOUGHT for world building ideas, but not specific to the current part of the story/narrative.
Anything that goes into writing or crafting needs to be captured in “real time” which means dumping it right in the middle of whatever you are doing.
I started with something more “contained” and easier to manage because actual users will go off script every chance they get, and this is basically like playing chess against yourself while reading a book on how to play chess. But, I may have found a kind of working compromise in terms of format and what needs to be captured. Will need a few days to see how it holds up, but right now, this is the basic idea:
Initial PROMPT to get the story started, followed by THOUGHTS that examine them from a gaming perspective, an ACTION, my THOUGHTS, another PROMPT, and.. this is where I was having a tough time because some of the mechanics were not being captured in the THOUGHTS prior. It was only as I wrote the PROMPT that I figured-out certain details or actions that needed to be in play. So when I write a PROMPT that contains these other elements, I write a LOGIC section below them to explain why I “prompted” the way I did.
In crafting the story as you go, the PROMPT is also part of the THOUGHT process! I’m sure anyone giving this a try will be writing and re-writing their prompt as part of the process. Having this extra LOGIC step seems to clean that up, but I don’t think any ML algo will ever keep track of story elements, have ideas on where to take the story next, and then backtrack. Perhaps the “prompt” is some adversarial output from the thoughts, but still internal to process, leading to more thoughts (aka the logic), which leads to the actual output.
Just my 2 cents.
Found a rhythm using PLAT (Prompt. Logic. Action. Thought.) but am only averaging 185 words per step. That would be about 18,000 words for 100 steps, or 54,000 words for 300 (which is the very bottom end of book territory). Agree that 100 steps is no story, but waiting to reach 100 steps before checking-in is waiting to long.
Would recommend anyone near the 20 step or 10 pages mark send that in for feedback before going further. I’m going to ignore my own advice because I’d like to complete the first 3 scenes, which is closer to 10% of the full story.
People are concerned about upfront time commitment, while also being focused on 100 step minimum. In another comment I went over how 250 steps works as a better minimum, but to keep all the numbers aligned, perhaps every story should be in 3 acts of 100 steps each (with at least 300 steps being a requirement; handing-in 299 steps would seem sloppy and rude). That would make each “short” story worth $6k, and each act $2k, which is the same 10% of 1,000 steps. Except, handing in the first act should only reserve your $6k payout, not result in getting $2k at a time (desire for finished products and not having to burden anyone with increased tracking/management). There could also be an $18k cap (for 3 short stories of at least 300 steps each) to both limit the number of short stories submitted and let people know there is no “dominating” of the short story space.
With no idea what the arc of the run/story will be, it’s really hard to plan for 3 acts, so maybe not so useful. But did want to leave another comment about scenes. With 4 scenes being about 50 steps, just as a reference, we can look at the number of scenes in a movie to figure each run could be 500 to 750 steps in total length. I just don’t see 1,000 steps as being anything other than an arbitrary dataset requirement. 250-300 steps as a playable run. 500 to 600 steps as a “movie length” representation. And then to double that?
The mental requirement to “film” a Lord of the Rings trilogy while also “filming” the behind the scenes of that filming and also “filming” the real-time documentary required to keep track of everything… while not being clear on how that extra “run time” translates into being better training data.
Is there going to be a “THIS” post, using sample work that you really like and “demanding” all other entries follow that exact format? How will variations in formatting be addressed? Does it need to be?
If you get something that checks all the right boxes, with one exception that leads to a rejection, I think we’d all like to know what that one must-have is.
Using scenes as a marker has some added benefit as I find myself leaving high level comments about some of the next scenes (I had nothing planned beyond the start, but the natural progression leads to speculation about future events or details). This is some of that looking ahead data that this project wanted to capture. Perhaps there should be a FUTURE keyword to wrap these things under? It would basically be a THOUGHT for world building ideas, but not specific to the current part of the story/narrative.
Anything that goes into writing or crafting needs to be captured in “real time” which means dumping it right in the middle of whatever you are doing.