[HN Gopher] Automating Interactive Fiction Logic Generation with...
       ___________________________________________________________________
        
       Automating Interactive Fiction Logic Generation with LLMs in Emacs
        
       Author : dskhatri
       Score  : 58 points
       Date   : 2025-03-31 15:57 UTC (7 hours ago)
        
 (HTM) web link (blog.tendollaradventure.com)
 (TXT) w3m dump (blog.tendollaradventure.com)
        
       | kleiba wrote:
       | What strikes me as odd in the video: why would the author not
       | fill the paragraphs?!
        
         | iLemming wrote:
         | Probably because then, it would make your paragraphs 'rigid',
         | whereas if you have visual-line-mode and don't truncate lines
         | -- the text would just wrap around -- you only need to adjust
         | the width of the window. That works nicely in "distraction-
         | free" modes like writeroom-mode.
         | 
         | I used to fill the paragraphs all the time, turns out -- it's
         | really better to leave them as they are, because you can never
         | get the satisfiable number of `fill-column` -- the default
         | works in some cases, for others, you'd want it to be wider,
         | etc.
        
       | spudlyo wrote:
       | GPTel is a very powerful interface for working with LLMs in
       | Emacs. It took me a while to understand that its real value isn't
       | what you get with M-x gptel, which creates a dedicated chat
       | session and buffer, but rather the ability to sling prompts,
       | context, and LLM output around in a native Emacs way. You can add
       | to the context from dired, from a file, from a buffer, you can
       | select from various prescribed system prompts for different
       | functionality, you can prompt from the minibuffer, the kill-ring,
       | the existing buffer, a selection, you can have the responses go
       | to the minibuffer, the kill-ring, a buffer, the echo area -- it's
       | extremely flexible.
       | 
       | I have a little helper function that uses gptel-request that I
       | use while reading Latin texts. It sets the system prompt so the
       | LLM acts as either a Latin to English translator, or with a
       | prefix argument it breaks down the grammatical structure and
       | vocabulary of a sentence for me. It's very cool.
        
       | IngoBlechschmid wrote:
       | Gwern shared an idea how to exploit the strength of current-
       | generation LLMs, despite their weaknesses, for "create your own
       | adventure"-style fiction. https://gwern.net/cyoa Having people
       | vote on AI-generated potential continuations should yield better
       | results and cut costs at the same time.
       | 
       | From the title I thought this was an implementation of Gwern's
       | idea, but it's not.
        
         | lawlessone wrote:
         | >Having people vote on AI-generated potential continuations
         | should
         | 
         | So the story never ends?
        
         | ianbicking wrote:
         | There have been experiments in wiki-style cyoa generation
         | (letting the public create options instead of an LLM), but they
         | suffer the same problem as LLM-generated stories: aimless
         | wandering and lack of consistency.
         | 
         | (As I think about it, an LLM generation should be thought of as
         | a many-author situation, as each generation comes in cold)
         | 
         | Stories need pacing, which exists over many passages, not just
         | at the choice level. And then the passages should all be based
         | on a single underlying world. Both of these fall apart quickly
         | without a guiding author.
         | 
         | I think this is resolvable with LLMs and appropriate prompting,
         | but the naive approach seems cool only until you actually play
         | out a few stories
        
       | zoogeny wrote:
       | This is one of the most promising uses of LLMs that I have found
       | in my own work. Many times I have an idea for a refactor or even
       | a feature but I have this mental reluctance just due to the
       | amount of code I would have to write. Like, I have this counter
       | in my head on the number of key-strokes it will take to write a
       | wrapper object in several places, and I hesitate.
       | 
       | Just being able to tell an LLM "rewrite all of this code using
       | this new pattern" and then dozens of code sites are correctly
       | updated is a huge help. It makes me consider bigger refactoring
       | or minor features that I might normally skip because I am lazy.
        
       ___________________________________________________________________
       (page generated 2025-03-31 23:00 UTC)