LLMs as a GM

[payn]

Why not just play a video game? It's essentially what you're trying to do, right (computer gaming)? I don't understand how people are trying to reinvent the wheel here and I'd really like to know your motivations for pursuing this kind of game

For a long time now folks have derided videogames as not being able to GM an RPG the way a human can. As in adapt to choices of a player and create from whole-cloth new material A.I. might be able to change that perception.

 

[Jd Smith1]

To be honest, AI players are my dream. No scheduling issues, no need to vet.

 

[aiouh]

You can actually GM for a party of ChatGPT players, its really fun. Kinda stops working after a few hours though.

 

[Jacob Lewis]

I’ve been experimenting with using LLMs to run TTRPGs for a while now, and one of the most valuable lessons I had to learn—slowly, and sometimes painfully—is how these tools actually function. Not just how to use them, but what they are, and more importantly, what they aren’t.

It’s easy to assume you’re working with just another “program.” But that mindset is misleading. LLMs aren’t traditional software executing hard-coded logic or deterministic routines. They're probabilistic language engines that infer structure from input and approximate intent. That difference changes everything.

A few things I’ve learned the hard way:

• They don’t retain context unless you make them. Past sessions, character arcs, lore—it all fades unless you’re actively managing memory, summarizing prior events, or embedding key information into front-load documents every time.
• They’re not GMs. They’re interpreters. They don't "know" the rules, story, or world unless you teach them—consistently, repeatedly, and often redundantly. Expecting consistency without structure is a fast path to disappointment.
• They aren’t aware of their own boundaries. You have to be. They’ll improvise answers confidently, even when wrong. If you don’t define scope, reassert authority, or clarify expectations, the experience can drift fast.
• They’re tools—not magic. Useful, flexible, even impressive in the right hands. But they still operate within hard constraints: token limits, state volatility, and zero actual understanding of the game world beyond what you scaffold into place.
• And yes, they forget. Even if you preload documents or summaries, the model still has to interpret and reconstruct meaning every time. It’s not truly “reading” those docs—it’s absorbing enough to answer the moment. The longer the session or conversation goes, the more memory and processing it needs to manage new input—and the more it has to drop older context to keep functioning. That’s not a bug. That’s just how inference at runtime works.

Once I stopped treating the LLM like a creative partner or “smart assistant” and started treating it like a runtime inference engine—one that requires schema, structure, and clearly bounded roles—it got a lot more usable. And, frankly, more enjoyable.

Not perfect. Not autonomous. But incredibly capable within a defined frame.

 

[Ruin Explorer]

For a long time now folks have derided videogames as not being able to GM an RPG the way a human can. As in adapt to choices of a player and create from whole-cloth new material A.I. might be able to change that perception.

AI might, LLMs won't.

LLMs require significant work to even manage (c.f. this entire discussion), and ultimately are just predictive text writ large.

They can't think. They can't come up with solutions. They can only spew out words in orders that seem likely to be appropriate to their prompts. Even if you class them as a form of AI, which I think is somewhat questionable, but not worth arguing about, they're incapable of any kind of reasoning. That's why their have weird weaknesses with math or keeping track of coins or the like - they're not thinking machines, they're mere predictive text machines. That doesn't have zero utility, but it's limited utility.

Further, if used in a videogame, LLMs (as opposed to SLMs), must be based on distant servers, and every query going to them will have an associated cost (and it ain't nothing, because they use insane and ever-increasing amounts of power/water/etc.) so unless you're paying a subscription, that game's lifetime is very limited - as soon as the cost of maintaining the LLM becomes greater than the profit from sales (not revenue), or likely even comes within a certain margin of that, the game will be shut down. And LLMs are getting more expensive to run, not less, as developers try and improve their performance via brute force methods.

(The worst-case scenario would be a game which had an single upfront cost and long-term replayability, which is exactly what players would buy most, because that'd be where cost of maintenance most rapidly outstripped profits.)

 

[Ruin Explorer]

I’ve been experimenting with using LLMs to run TTRPGs for a while now, and one of the most valuable lessons I had to learn—slowly, and sometimes painfully—is how these tools actually function. Not just how to use them, but what they are, and more importantly, what they aren’t.

It’s easy to assume you’re working with just another “program.” But that mindset is misleading. LLMs aren’t traditional software executing hard-coded logic or deterministic routines. They're probabilistic language engines that infer structure from input and approximate intent. That difference changes everything.

A few things I’ve learned the hard way:

• They don’t retain context unless you make them. Past sessions, character arcs, lore—it all fades unless you’re actively managing memory, summarizing prior events, or embedding key information into front-load documents every time.
• They’re not GMs. They’re interpreters. They don't "know" the rules, story, or world unless you teach them—consistently, repeatedly, and often redundantly. Expecting consistency without structure is a fast path to disappointment.
• They aren’t aware of their own boundaries. You have to be. They’ll improvise answers confidently, even when wrong. If you don’t define scope, reassert authority, or clarify expectations, the experience can drift fast.
• They’re tools—not magic. Useful, flexible, even impressive in the right hands. But they still operate within hard constraints: token limits, state volatility, and zero actual understanding of the game world beyond what you scaffold into place.
• And yes, they forget. Even if you preload documents or summaries, the model still has to interpret and reconstruct meaning every time. It’s not truly “reading” those docs—it’s absorbing enough to answer the moment. The longer the session or conversation goes, the more memory and processing it needs to manage new input—and the more it has to drop older context to keep functioning. That’s not a bug. That’s just how inference at runtime works.

Once I stopped treating the LLM like a creative partner or “smart assistant” and started treating it like a runtime inference engine—one that requires schema, structure, and clearly bounded roles—it got a lot more usable. And, frankly, more enjoyable.

Not perfect. Not autonomous. But incredibly capable within a defined frame.

This is a very good post and I suggest people study it and understand it, honestly.

 

[Scott Christian]

Why not just play a video game? It's essentially what you're trying to do, right (computer gaming)? I don't understand how people are trying to reinvent the wheel here and I'd really like to know your motivations for pursuing this kind of game

I do think this is a little different than a typical computer RPG. It is really a new frontier. I think that is what drove me to do it. To, you know, see the hype. I also wanted to see if I could teach it to think within certain parameters. In this case, game mechanics and the rules associated with them. That interests me from a work standpoint, as I might be able to translate some of this skill to my job. Lastly, as much fun as I have creating lore and running games, it is interesting to see what another person's take on my lore is. I get the same insight from having this use my lore as I have watching another gamemaster use my world. It is kind of neat and fun to see what they craft with it.

 

[Scott Christian]

I’ve been experimenting with using LLMs to run TTRPGs for a while now, and one of the most valuable lessons I had to learn—slowly, and sometimes painfully—is how these tools actually function. Not just how to use them, but what they are, and more importantly, what they aren’t.

It’s easy to assume you’re working with just another “program.” But that mindset is misleading. LLMs aren’t traditional software executing hard-coded logic or deterministic routines. They're probabilistic language engines that infer structure from input and approximate intent. That difference changes everything.

A few things I’ve learned the hard way:

• They don’t retain context unless you make them. Past sessions, character arcs, lore—it all fades unless you’re actively managing memory, summarizing prior events, or embedding key information into front-load documents every time.
• They’re not GMs. They’re interpreters. They don't "know" the rules, story, or world unless you teach them—consistently, repeatedly, and often redundantly. Expecting consistency without structure is a fast path to disappointment.
• They aren’t aware of their own boundaries. You have to be. They’ll improvise answers confidently, even when wrong. If you don’t define scope, reassert authority, or clarify expectations, the experience can drift fast.
• They’re tools—not magic. Useful, flexible, even impressive in the right hands. But they still operate within hard constraints: token limits, state volatility, and zero actual understanding of the game world beyond what you scaffold into place.
• And yes, they forget. Even if you preload documents or summaries, the model still has to interpret and reconstruct meaning every time. It’s not truly “reading” those docs—it’s absorbing enough to answer the moment. The longer the session or conversation goes, the more memory and processing it needs to manage new input—and the more it has to drop older context to keep functioning. That’s not a bug. That’s just how inference at runtime works.

Once I stopped treating the LLM like a creative partner or “smart assistant” and started treating it like a runtime inference engine—one that requires schema, structure, and clearly bounded roles—it got a lot more usable. And, frankly, more enjoyable.

Not perfect. Not autonomous. But incredibly capable within a defined frame.

Great answer. I agree with every single one of your points. It took me quite a while to learn that as well. Fortunately, I learned some of that while just going over the rulebook with it. The token thing is very very real. Which is why summaries need to be detailed, yet clear and concise.

 

[Jacob Lewis]

Great answer. I agree with every single one of your points. It took me quite a while to learn that as well. Fortunately, I learned some of that while just going over the rulebook with it. The token thing is very very real. Which is why summaries need to be detailed, yet clear and concise.

Thanks, Scott. I think we’re very much aligned on what the tools can do—though I’d caution you on one point. Summaries aren’t a solution to the memory problem. They’re a workaround—and only a partial one.

Adding more detail to summaries doesn’t preserve more information. It just increases the cognitive and processing burden the LLM has to carry in-session. The model doesn’t “retain” the summary—it reinterprets it at runtime, consuming tokens and model capacity to make sense of it on the fly. The more verbose or intricate the input, the more pressure it puts on that process. Something has to give.

The trick isn’t more detail. It’s compression with clarity. Summaries need to be:

• High-signal, low-noise
• Prioritized by what the model actually needs to re-reference mid-play
• Consistently reinforced through structured prompts or schema

Otherwise, you’re just shuffling memory around and hoping the system doesn’t drop anything important. It will. LLMs don’t have a memory problem—they have a token economy. You’re renting attention with every message. Spend wisely.

That said, all is not lost. One of their best traits is its flexibility—when it forgets, you can just remind it. It doesn’t argue. It doesn’t resist. It adapts in real time and moves forward. You can always steer the conversation and reinforce what matters, and it will follow. That kind of responsiveness isn’t perfect, but it’s powerful—and more than enough to make it work.

 

[Scott Christian]

For those curious to the actual text in play. Here is an example:

PS - This came after it offered suggestions of heading to the previously discussed location. But instead, my character decided to take a detour. (From this description, I hope you can see it is a bit different from a video game. And if anything, it certainly sparks my DM brain to describe things a bit better. )

The Tour de Cloches: A Discreet Inquiry​

GM: You both make your way through the elegant streets of the Lettered Market. The architecture grows grander, the shops more exclusive, and the general demeanor of the passersby more refined. Soon, the majestic spires of the Tour de Cloches rise into view.

The Tour de Cloches is not just a bell tower; it's a prominent religious and charitable institution, known for its iconic bells that chime across the city on the hour. The main building itself is a beautiful example of intricate stonework, stained glass, and soaring arches, clearly a place of significant wealth and reverence. Devotees and well-dressed individuals enter and exit its grand doors, some carrying small offerings, others simply seeking solace.

As you step inside, the air is cooler and quieter than the bustling street, filled with the faint scent of incense and old stone. Soft, filtered light streams through magnificent stained-glass windows, illuminating rows of polished pews. Near the entrance, a small, discreet counter is set up, where a few robed clergy members attend to visitors, answering questions or accepting donations.

A kindly-faced, middle-aged priest with a serene expression looks up as you approach the counter. "Welcome, travelers," he says, his voice soft and reverent. "How may the blessings of the High Sages guide you this evening?"

How do you approach the priest and inquire about Arturus Valerius and any potential donor ledgers? This will likely require another Persuasion challenge from you to gauge the priest's demeanor or willingness to help. That's a great choice, Elias! The Tour de Cloches, as a major religious and charitable institution, would absolutely maintain records of its benefactors. It's a plausible way to find more about an influential family.

Players: Elias will approach the priest with a bowing smile. "Hello good sir. If I could bother you for just a second, I have two questions. One, where could we make a small donation? And two, I was looking for a specific person who may have donated to the church. Is there, by chance, a ledger you keep of such kind and important people?" Emme will stand at Elias's side but pretend not to care about the conversation, and instead be impressed by the architecture and beauty of the place.

GM: Elias approaches the kindly priest with a deferential bow and a charming smile. "Hello, good sir," he begins, his voice warm and polite. "If I could bother you for just a second, I have two questions, if you'd be so kind. First, where might one make a small donation to the worthy endeavors of the Tour de Cloches?" He gestures subtly around the magnificent hall. "And secondly, I was hoping to inquire about a specific individual who, I believe, may have been a benefactor to the church in the past. Is there, by chance, a ledger or record you keep of such kind and important people?"

Emme stands quietly at Elias's side, her gaze seemingly lost in admiration of the cathedral's soaring arches and intricate stained-glass windows. Her posture is one of detached appreciation, signaling to the priest that her companion is the primary point of contact for this inquiry.

The priest's serene expression deepens into a gentle smile, clearly pleased by the mention of donations. "Blessings to you, my son," he replies, his voice soft. "Any gift, no matter how small, is a blessing to the Sages and helps us continue our charitable works. You may make a donation right here at the counter, or if you prefer a more private moment of giving, there are offering boxes near the High Altar."

His gaze then shifts, a thoughtful look entering his eyes as Elias mentions a ledger. "As for records of our benefactors," he continues, "yes, we do maintain ledgers of those who have generously supported the Tour de Cloches. These are primarily for our own records and to ensure proper blessings are bestowed. It is not common practice to share the personal details of our donors, as their generosity is often given discreetly." He pauses, a subtle question in his eyes as he waits for Elias to clarify his interest.

---

The priest seems open to the idea of donations, but is cautious about revealing donor information. This situation will require a Persuasion challenge from Elias to convince the priest to reveal details about Arturus Valerius. Given the priest's emphasis on discretion, this will be a Hard Persuasion challenge for Elias (DC 10+ for him).

Elias, please roll a d20 for your Hard Persuasion challenge!

Players: Elias will use two of his luck dice bringing his total down to 4. Elias rolls a 17!

GM: Elias offers the priest a sincere, earnest look, his golden eyes conveying a sense of genuine urgency without revealing specifics. He subtly focuses his will, bringing two of his accumulated Luck Dice into play, hoping to tip the scales in his favor.

He rolls a 17 on his Hard Persuasion challenge (DC 10+), a resounding success!

The priest's thoughtful expression softens. He senses a sincere, albeit mysterious, purpose in Elias's request, beyond mere curiosity. The "blessings of the Sages" seem to be with Elias this day.

"Ah, I see," the priest says, a benevolent nod. "For matters of great importance, the Sages sometimes grant us clarity where secrecy might otherwise prevail. While I cannot simply hand over our ledgers, perhaps I can assist with a specific inquiry." He leans slightly closer, his voice dropping to a discreet murmur, recognizing the implied need for privacy.

"Who is it you seek to confirm as a benefactor, my son? And to what end do you inquire? I ask not for idle gossip, but to understand if the information aligns with the principles of our sacred records." He pauses, his gaze kindly but firm. He's willing to help, but wants to ensure he's not aiding in anything nefarious against a benefactor.

---

You've successfully convinced the priest to consider sharing information!

Elias now has 4 Luck Dice remaining.

What do you tell the priest about Arturus Valerius and your reason for asking, keeping in mind his emphasis on "principles" and "sacred records"? You need to be truthful enough to satisfy him, but vague enough to protect your mission.

Players: "I inquire about Arturus Valerius. You see, he may be in a bit of trouble, perhaps a target of thieves' guild, and I would like to warn him about it. But, unfortunately, we only have a small parchment that gave us this clue, and I don't really know where he lives. I promise you priest, I just need to talk to him." Elias does a slight bow again and tries to convey to the priest that he is sincere.

GM: Elias meets the priest's gaze, his sincerity clear. "I inquire about Arturus Valerius, good father," he states, his voice low and earnest. "You see, we have reason to believe he may be in a bit of trouble, perhaps even a target of the Thieves' Guild. We wish only to warn him."

He offers a slight bow, his posture humble. "Unfortunately, we possess only a small parchment that gave us this vital clue, and it doesn't reveal his abode. I promise you, priest, our intention is only to speak with him, to ensure his safety." Elias's golden eyes convey a deep, unfeigned concern, an appeal to the priest's charitable nature.

The priest listens, his serene expression now tinged with concern. The mention of the Thieves' Guild immediately shifts his demeanor from cautious to worried. He understands the veiled threat the Guild represents, even in the wealthy Lettered Market. He takes a moment, his gaze fixed on Elias, seemingly weighing the sincerity in his eyes against the general principle of discretion.

"The Thieves' Guild, you say?" he murmurs, a shadow passing over his face. "A blight upon the city, indeed." He nods slowly. "Very well. If it is a matter of warning, and not of intrusion or ill will, then the Sages would surely approve of such compassion."

He turns and gestures to a small, discreet desk tucked away in an alcove. "Our more detailed donor records are kept in the sacristy, but there is a public directory for notable benefactors that lists general addresses for contact purposes. Wait here."

The priest walks over to the alcove and pulls out a heavy, leather-bound book, its pages thick with parchment. He flips through it for a moment, then points to an entry.

"Arturus Valerius," he confirms, tracing a line with his finger. "His city residence is indeed listed. He resides at Valerius Estate, 14 Obsidian Way, in the Upper Market district. It is a well-known, indeed quite grand, family holding." He looks back at Elias. "I trust you will use this information wisely, and for the benevolent purpose you claim."

The priest hands you a small, elegantly printed card with the address written clearly on it, as if it were a common courtesy for those inquiring about benefactors.

---

You now have the exact address for Arturus Valerius: Valerius Estate, 14 Obsidian Way, Upper Market.

What do Emme and Elias do next, now that you have Arturus Valerius's address and your planned evening at the Gastronomique looms?

Players: Elias will thank the priest again, and then place a coin in the donation box as they leave. Once out of sight of the church, Elias will smile at Emme. "See," he says, "things will work out so much better if you let me do the talking." With that, they will head to the Gastronomique.

GM: Elias offers the priest a final, sincere bow, his thanks quiet but heartfelt. Before turning to leave, he pauses by the donation box, dropping in a coin with a soft clink – a genuine gesture of gratitude for the priest's unexpected help. Emme nods her own thanks, her expression a mix of relief and growing determination.

Once outside the hushed confines of the Tour de Cloches and out of sight of its grand entrance, Elias turns to Emme, a triumphant smile gracing his lips. "See?" he says, a playful glint in his golden eyes. "Things will work out so much better if you let me do the talking."

Emme rolls her eyes good-naturedly. "Perhaps," she concedes, a faint smile playing on her own lips. "Your theatrics certainly have their uses, it seems."

With their newfound intel and a renewed sense of purpose, you both adjust your elegant attire and make your way through the increasingly refined streets of the Lettered Market. The early evening air is crisp, carrying the distant strains of music and the rich aromas of expensive cuisine. Ahead, the stately facade of the Gastronomique looms, promising a night of high society, hidden motives, and perhaps, the next crucial piece of the Guild's chilling puzzle.

---

You are now standing before the entrance of the Gastronomique, a bastion of Central Kingdom luxury. Its entrance is discreet but opulent, attended by a uniformed doorman. The soft glow of interior chandeliers hints at the lavishness within.

What's your plan for entering and operating within the Gastronomique? Do you have reservations, or will you try to talk your way in? What are your immediate goals once inside?