31–40 of 46 posts
Sorry, but that isn't a verb recognition problem, it's a comprehension problem. I agree with the downvoted poster that LLM integration would significantly improve the end user experience. However the LLM should not be arbiter of game state (as they suggested), but simply the translator that ensures that the players instructions are understood by the game.
I was there in the 1980s; writing your own Infocom-style text adventure games was a common project among my peers. There are probably lots of unfinished (or even finished-but-unshared) games out there on old floppy disks in closets. I have a couple of my own, now archived on my home server.
Willing to share them?
LLMs can enhance text adventures. I'm not saying having LLMs narrate the entire situation. I'm saying have the LLM sit between gamestate and the player. The LLM is the UI. Essentially the LLM can see the current game state and possible moves and it's the LLMs job to change the game state and report the current game state to the user (via a well written narrative). That keeps the world consistent and structured, but t…
Even Inform6 with the English library running under a Pentium would run circles over llm's. They aren't even close. If7 will curb-stomp it. NPC's running around were a thing even in the 80's, see The Hobbit. That on a ZX spectrum. 8 bit CPU, 48k of RAM. With if6 and Zmachine games for 16 bit and 32 bit computers in the 90's llm's can't even compete. From Jigsaw, Anchorhead, Curses, Spider and Web to that anagram word…
https://drive.google.com/file/d/0Bxv0SsvibDMTYkFJbUswOHFQclE... [PDF]
Earlier quoted context omitted.
Even Inform6 with the English library running under a Pentium would run circles over llm's. They aren't even close. If7 will curb-stomp it. NPC's running around were a thing even in the 80's, see The Hobbit. That on a ZX spectrum. 8 bit CPU, 48k of RAM. With if6 and Zmachine games for 16 bit and 32 bit computers in the 90's llm's can't even compete. From Jigsaw, Anchorhead, Curses, Spider and Web to that anagram word…
How? The LLM is in agentic mode and pulling and pushing the switches of the game. I don't see how these engines can curb stomp it.
An Inform 6 compiler can run under DOS/386 or Linux/BSD for 486 and compile a really large and functional game in seconds. For Inform7, maybe a Pentium II/III and a bunch of seconds too.
An AI to do the same, not even a 'modern' Core Duo with a GL 2.1 adapter (the lowest of the lowest bearable specs for light web browsing and office work) can't even run a consistent world.
I spent a bit of time using PAWS on an amstrad pcw to pen my own text adventure masterpiece. It's probably still down there packed up in the cellar .... Maybe I should dig it out again.
Earlier quoted context omitted.
There have been attempts, the results were not promising: https://intfiction.org/t/first-full-game-available-on-new-ll...
It looks as if the problem for this one was a game design issue. They had the LLM be sarcastic and ignore well known adventure game prompts on purpose. It's an easy fix to make the LLM more obedient and polite. The biggest issue is attempts to hack the LLM, to get at hidden gamestate. But I feel this can be easily remedied by just not providing the LLM with hidden game state.
That would keep its behavior passive and restricted to simply the current state, since it would retain no memory of previous actions.