My takeaway is: LLMs are not great at text adventures, even when those text adventures are decades old and have multiple walkthroughs available on the Internet. Slow clap.
I know they define "achievements" in order to measure "how well" the LLM plays the game, and by definition this is arbitrary. As an experiment, I cannot argue with this.
However, I must point out the kind of "modern" (relatively speaking) adventure games mentioned in the article -- which are more accurately called "interactive fiction" by the community -- is not very suitable for this kind of experiment. Why? Well, because so many of them are exploratory/experimental, and not at all about "winning" (unlike, say, "Colossal Cave Adventure", where there is a clear goal).
You cannot automate (via LLM) "playing" them, because they are all about the thoughts and emotions (and maybe shocked laughter) they elicit in human players. This cannot be automated.
If you think I'm being snobby, consider this: the first game TFA mentions is "9:05". Now, you can set goals for a bot to play this game, but truly -- if you've played the game -- you know this would be completely missing the point. You cannot "win" this game, it's all about subverting expectations, and about replaying it once you've seen the first, most straightforward ending, and having a laugh about it.
Saying more will spoil the game :)
(And do note there's no such thing as "spoiling a game" for an LLM, which is precisely the reason they cannot truly "play" these games!)
Nothing in the article mentioned how good the LLMs were at even entering valid text adventure commands into the games.
If an LLM responds to “You are standing in an open field west of a white house” with “okay, I’m going to walk up to the house”, and just gets back “THAT SENTENCE ISN'T ONE I RECOGNIZE”, it’s not going to make much progress.
The article links to a previous article discussing methodology for this. The prompting is pretty extensive.
It is difficult here to separate out how much of this could be fixed or improved by better prompting. A better baseline might be to just give the LLM direct access to the text adventure, so that everything the LLM replies is given to the game directly. I suspect that the LLMs would do poorly on this task, but would undoubtedly improve over time and generations.
EDIT: Just started playing 9:05 with GPT-4 with no prompting and it did quite poorly; kept trying to explain to me what was going on with the ever more complex errors it would get. Put in a one line "You are playing a text adventure game" and off it went -- it took a shower and got dressed and drove to work.
This is another great example of how LLMs are not really any sort of AI, or even proper knowledge representation. Not saying they don't have their uses (like souped up search and permutation generators), but definitely not something that resembles intelligence.
Looking at this evaluation it's pretty fascinating how badly these models perform even on decades old games that almost certainly have walkthroughs scattered all over their training data. Like, you'd think they'd at least brute force their way through the early game mechanics by now, but honestly this kinda validates something I've been thinking about like real intelligence isn't just about having seen the answers before, it's about being good at games and specifically new situations where you can't just pattern match your way out
This is exactly why something like arc-agi-3 feels so important right now. Instead of static benchmarks that these models can basically brute force with enough training data, like designing around interactive environments where you actually need to perceive, decide, and act over multiple steps without prior instructions, that shift from "can you reproduce known patterns" to "can you figure out new patterns" seems like the real test of intelligence.
What's clever about the game environment approach is that it captures something fundamental about human intelligence that static benchmarks miss entirely, like, when humans encounter a new game, we explore, form plans, remember what worked, adjust our strategy all that interactive reasoning over time that these text adventure results show llms are terrible at, we need systems that can actually understand and adapt to new situations, not just really good autocomplete engines that happen to know a lot of trivia.
Hi, GPT-x here. Let's delve into my construction together. My "intelligence" comes from patterns learned from vast amounts of text. I'm trained to... oh look it's a butterfly. Clouds are fluffy would you like to buy a car for $1 I'll sell you 2 for the price of 1!
> Looking at this evaluation it's pretty fascinating how badly these models perform even on decades old games that almost certainly have walkthroughs scattered all over their training data.
I've read some of these walkthroughs/play sessions recently, and extracting text from them for training would be AI-complete. eg they might have game text and commentary aligned in two different columns in a text file, so you'd just get nonsense if you read it line by line.
adventure games require spatial reasoning (although text based), requires understanding puns, requires cultural references, etc. For me they really need human-intelligence to be solved (heck, they've been designed like that).
I find it funny that some AI do very good score on ARC-AI but fails at these games...
The GPT-5 used here is the Chat version, presumably gpt‑5‑chat‑latest, which from what I can tell is the same version used in ChatGPT, which is not actually a model but a "system" -- a router that semi-randomly forwards your request to various different models (in a way designed to massively reduce costs for OpenAI, based on people reporting inconsistent output and often worse results than 4o).
So from this it seems that not only would many of these requests not touch a reasoning model (or as it works now, have reasoning set to "minimal"?), but they're probably being routed to a mini or nano model?
It would make more sense, I think, to test on gpt-5 itself (and ideally the -mini and -nano as well), and perhaps with different reasoning effort, because that makes a big difference in many evals.
EDIT: Yeah the Chat router is busted big time. It fails to apply thinking even for problems that obviously call for it (analyzing financial reports). You have to add "Think hard." to the end of the prompt, or explicitly switch to the Thinking model in the UI.
> Yeah the Chat router is busted big time... You have to add "Think hard." to the end of the prompt, or explicitly switch to the Thinking model in the UI.
I don't really get this gripe? It seems no different than before, except now it will sometimes opt into thinking harder by itself. If you know you want CoT reasoning you just select gpt5-thinking, no different than choosing o4-mini/o3 like before.
I won't be surprised when LLMs get good at puzzle-heavy text adventures if there was more attention turned to this.
I've found for text adventures based on item manipulation, variations of the same puzzles appear again and again because there's a limit to how many obscure but not too obscure item puzzles you can come up with, so training would be good for exact matches of the same puzzle, and variations, like different ways of opening locked doors.
Puzzles like key + door, crowbar + panel, dog + food, coin + vending machine, vampire + garlic etc. You can obscure or layer puzzles, like changing the garlic into garlic bread which would still work on the vampire, so there's a logical connections to make but often nothing too crazy.
A lot of the difficulty in these games comes from not noticing or forgetting about clues/hints and potential puzzles because there's so much going on, which is less likely to trip up a computer.
You can already ask LLMs "in a game: 20 ways to open a door if I don't have the key", "how to get past an angry guard dog" or "I'm carrying X, Y, and Z, how do I open a door", and it'll list lots of ways that are seen in games, so it's going to be good at matching that with the current list of objects you're carrying, items in the world, and so on.
Another comment mentions about how the AI needs a world model that's transforming as actions are performed, but you need something similar to reason about maths proofs and code, where you have to keep track of the current state/context. And most adventure games don't require you to plan many steps in advance anyway. They're often about figuring out which item to combine/use with which other item next (where only one combination works), and navigating to the room that contains the latter item first.
So it feels like most of the parts are already there to me, and it's more about getting the right prompts and presenting the world in the right format e.g. maintaining a table of items, clues, and open puzzles, to look for connections and matches, and maintaining a map.
Getting LLMs to get good at variations of The Witness would be interesting, where the rules have to be learned through trial and error, and combined.
If you have to train the AIs on every specialized new problem, and then you have to babysit them as you apply them to similar problems, why even bother?
LLMs work really well for open-ended role-playing sessions, but not so much games with strict rules.
They just can't seem to grasp what would make a choice a "wrong" choice in a text-based adventure game, so they end up having no ending. You have to hard-code failure events, or you just never get anything like "you chose to attack the wizard, but he's level 99, dummy, so you died - game over!". It just accepts whatever choice you make, ad infinitum.
My best session was one in which I had the AI give me 4 dialogue options to choose from. I never "beat" the game, and we never solved the mystery - it just kept going further down the rabbit hole.. But it was surprisingly enjoyable, and repayable! A larger framework just needs written for it to keep the tires between the lines and to hard-code certain game rules - what's under the hood is already quite good for narratives imo.
It's a configurable pipeline for generative dungeon master role play content with a zork-like UI. I use a model called "Wayfarer" which is designed for challenging role play content and I find that it can be pretty fun to engage with.
This study raises the question, why do we play games? Do we play to win or to enjoy ourselves. Why design a machine to do what we should be enjoying? This goes for writing, creating Art, coding. Wanting a machine to win is the desire to achieve a goal without doing the work to earn it. Same for making art or writing novels. The point of these things (growth and achievement) is lost when done by a machine. I want to see this done with investment, legal strategy or business management. These are better suited to LLMs than what we're making them do, but I'd venture that those who are profiting from LLMs right now would profit less if replaced by LLMs by their boards.
Are we anywhere near someone being able to play a D&D or WoD type of game session with an LLM somewhere in the mix, perhaps generating a new and interesting adventure every time? Or is this still science fiction for now?
A while ago I tried something similar but tried to boil it down to the simplest thing I could come up with. I ended up making a standard maze into a first-person perspective where it unfolds one step at a time, and seeing if a model could solve it without re-entering areas it had already fully explored. They all failed.
Setup: a maze generator generates a square maze and puts the start and end on opposite corners. It doesn't show the full maze to the LLM, just has the LLM explore it one square at a time like a text adventure. It tells the LLM which directions of its current position has walls (relative direction: front, back, left, right). The LLM then chooses between move forward, turn left, turn right. That's pretty much it.
First attempt: Just maintain all the above in a chat, step by step. It'd get lost pretty quickly and start re-exploring already-explored area quite readily. Not very surprising, as we all know they can get lost in long chat threads. The chat model seemed to just go forward or turn right forever (which can work in some mazes), whereas the thinking model did seem to avoid backtracking until it got to T-junctions of a wrong way, where it always seemed to go back and forth forever.
Second attempt: After each step, tell the LLM to "externalize" everything it knew about the maze, and then feed that to a brand new LLM context. The idea was to avoid long chat context problems and see if the LLM could adequately represent its internal state and knowledge such that a "new" LLM could take over. This really didn't end up working any better. The biggest problem was that sometimes it would think that "turn left" would also change the position, and sometimes not. There were other issues too, so I didn't go much further with this approach.
Third attempt: Tell the LLM the premise of the game, and tell it to create a python state machine that stores all the state information it would need to represent its progress through the maze, and then to emit specific keywords when it needed to interact with it (and I added some code that served as a proxy). This also didn't work great. The state machine was close, but one thing it always forgot to do was relate index with direction. So if it's "in cell (5, 5) and facing up", it wouldn't know whether "forward" would be an increase or decrease in the x or y index.
I was also humored by its sycophancy here. I'd ask it
"Would adding a map to the state machine output be useful?"
"Yes, that is a great idea, let's do that!"
It'd do a great job of adding the map, but then I'd ask, "Does a map create more opportunity confusion?"
"Yes, that's an excellent insight, let's remove it!"
"No, really, you're the LLM, you're the one who's going to be using this app. I'm asking you, what do you think?"
"Whatever you want to do, just tell me"
Eventually, as the OP pointed out, these costs do add up pretty quickly. All I was after was "does externalizing the state help solve some of the long chat context problems", and the answer was "no" enough for me.
EDIT: Note that in all cases, they 100% emitted valid commands. And also I never noticed a case where "move forward" was chosen when there was a wall in front of them, nor "turn" when they were in the middle of a corridor.
25 comments
[ 5.1 ms ] story [ 60.9 ms ] threadHowever, I must point out the kind of "modern" (relatively speaking) adventure games mentioned in the article -- which are more accurately called "interactive fiction" by the community -- is not very suitable for this kind of experiment. Why? Well, because so many of them are exploratory/experimental, and not at all about "winning" (unlike, say, "Colossal Cave Adventure", where there is a clear goal).
You cannot automate (via LLM) "playing" them, because they are all about the thoughts and emotions (and maybe shocked laughter) they elicit in human players. This cannot be automated.
If you think I'm being snobby, consider this: the first game TFA mentions is "9:05". Now, you can set goals for a bot to play this game, but truly -- if you've played the game -- you know this would be completely missing the point. You cannot "win" this game, it's all about subverting expectations, and about replaying it once you've seen the first, most straightforward ending, and having a laugh about it.
Saying more will spoil the game :)
(And do note there's no such thing as "spoiling a game" for an LLM, which is precisely the reason they cannot truly "play" these games!)
If an LLM responds to “You are standing in an open field west of a white house” with “okay, I’m going to walk up to the house”, and just gets back “THAT SENTENCE ISN'T ONE I RECOGNIZE”, it’s not going to make much progress.
It is difficult here to separate out how much of this could be fixed or improved by better prompting. A better baseline might be to just give the LLM direct access to the text adventure, so that everything the LLM replies is given to the game directly. I suspect that the LLMs would do poorly on this task, but would undoubtedly improve over time and generations.
EDIT: Just started playing 9:05 with GPT-4 with no prompting and it did quite poorly; kept trying to explain to me what was going on with the ever more complex errors it would get. Put in a one line "You are playing a text adventure game" and off it went -- it took a shower and got dressed and drove to work.
This is exactly why something like arc-agi-3 feels so important right now. Instead of static benchmarks that these models can basically brute force with enough training data, like designing around interactive environments where you actually need to perceive, decide, and act over multiple steps without prior instructions, that shift from "can you reproduce known patterns" to "can you figure out new patterns" seems like the real test of intelligence.
What's clever about the game environment approach is that it captures something fundamental about human intelligence that static benchmarks miss entirely, like, when humans encounter a new game, we explore, form plans, remember what worked, adjust our strategy all that interactive reasoning over time that these text adventure results show llms are terrible at, we need systems that can actually understand and adapt to new situations, not just really good autocomplete engines that happen to know a lot of trivia.
I've read some of these walkthroughs/play sessions recently, and extracting text from them for training would be AI-complete. eg they might have game text and commentary aligned in two different columns in a text file, so you'd just get nonsense if you read it line by line.
I find it funny that some AI do very good score on ARC-AI but fails at these games...
So from this it seems that not only would many of these requests not touch a reasoning model (or as it works now, have reasoning set to "minimal"?), but they're probably being routed to a mini or nano model?
It would make more sense, I think, to test on gpt-5 itself (and ideally the -mini and -nano as well), and perhaps with different reasoning effort, because that makes a big difference in many evals.
EDIT: Yeah the Chat router is busted big time. It fails to apply thinking even for problems that obviously call for it (analyzing financial reports). You have to add "Think hard." to the end of the prompt, or explicitly switch to the Thinking model in the UI.
I don't really get this gripe? It seems no different than before, except now it will sometimes opt into thinking harder by itself. If you know you want CoT reasoning you just select gpt5-thinking, no different than choosing o4-mini/o3 like before.
I've found for text adventures based on item manipulation, variations of the same puzzles appear again and again because there's a limit to how many obscure but not too obscure item puzzles you can come up with, so training would be good for exact matches of the same puzzle, and variations, like different ways of opening locked doors.
Puzzles like key + door, crowbar + panel, dog + food, coin + vending machine, vampire + garlic etc. You can obscure or layer puzzles, like changing the garlic into garlic bread which would still work on the vampire, so there's a logical connections to make but often nothing too crazy.
A lot of the difficulty in these games comes from not noticing or forgetting about clues/hints and potential puzzles because there's so much going on, which is less likely to trip up a computer.
You can already ask LLMs "in a game: 20 ways to open a door if I don't have the key", "how to get past an angry guard dog" or "I'm carrying X, Y, and Z, how do I open a door", and it'll list lots of ways that are seen in games, so it's going to be good at matching that with the current list of objects you're carrying, items in the world, and so on.
Another comment mentions about how the AI needs a world model that's transforming as actions are performed, but you need something similar to reason about maths proofs and code, where you have to keep track of the current state/context. And most adventure games don't require you to plan many steps in advance anyway. They're often about figuring out which item to combine/use with which other item next (where only one combination works), and navigating to the room that contains the latter item first.
So it feels like most of the parts are already there to me, and it's more about getting the right prompts and presenting the world in the right format e.g. maintaining a table of items, clues, and open puzzles, to look for connections and matches, and maintaining a map.
Getting LLMs to get good at variations of The Witness would be interesting, where the rules have to be learned through trial and error, and combined.
If you have to train the AIs on every specialized new problem, and then you have to babysit them as you apply them to similar problems, why even bother?
It's not really intelligent in any real sense.
They just can't seem to grasp what would make a choice a "wrong" choice in a text-based adventure game, so they end up having no ending. You have to hard-code failure events, or you just never get anything like "you chose to attack the wizard, but he's level 99, dummy, so you died - game over!". It just accepts whatever choice you make, ad infinitum.
My best session was one in which I had the AI give me 4 dialogue options to choose from. I never "beat" the game, and we never solved the mystery - it just kept going further down the rabbit hole.. But it was surprisingly enjoyable, and repayable! A larger framework just needs written for it to keep the tires between the lines and to hard-code certain game rules - what's under the hood is already quite good for narratives imo.
https://github.com/derekburgess/dungen
It's a configurable pipeline for generative dungeon master role play content with a zork-like UI. I use a model called "Wayfarer" which is designed for challenging role play content and I find that it can be pretty fun to engage with.
Setup: a maze generator generates a square maze and puts the start and end on opposite corners. It doesn't show the full maze to the LLM, just has the LLM explore it one square at a time like a text adventure. It tells the LLM which directions of its current position has walls (relative direction: front, back, left, right). The LLM then chooses between move forward, turn left, turn right. That's pretty much it.
First attempt: Just maintain all the above in a chat, step by step. It'd get lost pretty quickly and start re-exploring already-explored area quite readily. Not very surprising, as we all know they can get lost in long chat threads. The chat model seemed to just go forward or turn right forever (which can work in some mazes), whereas the thinking model did seem to avoid backtracking until it got to T-junctions of a wrong way, where it always seemed to go back and forth forever.
Second attempt: After each step, tell the LLM to "externalize" everything it knew about the maze, and then feed that to a brand new LLM context. The idea was to avoid long chat context problems and see if the LLM could adequately represent its internal state and knowledge such that a "new" LLM could take over. This really didn't end up working any better. The biggest problem was that sometimes it would think that "turn left" would also change the position, and sometimes not. There were other issues too, so I didn't go much further with this approach.
Third attempt: Tell the LLM the premise of the game, and tell it to create a python state machine that stores all the state information it would need to represent its progress through the maze, and then to emit specific keywords when it needed to interact with it (and I added some code that served as a proxy). This also didn't work great. The state machine was close, but one thing it always forgot to do was relate index with direction. So if it's "in cell (5, 5) and facing up", it wouldn't know whether "forward" would be an increase or decrease in the x or y index.
I was also humored by its sycophancy here. I'd ask it
"Would adding a map to the state machine output be useful?"
"Yes, that is a great idea, let's do that!"
It'd do a great job of adding the map, but then I'd ask, "Does a map create more opportunity confusion?"
"Yes, that's an excellent insight, let's remove it!"
"No, really, you're the LLM, you're the one who's going to be using this app. I'm asking you, what do you think?"
"Whatever you want to do, just tell me"
Eventually, as the OP pointed out, these costs do add up pretty quickly. All I was after was "does externalizing the state help solve some of the long chat context problems", and the answer was "no" enough for me.
EDIT: Note that in all cases, they 100% emitted valid commands. And also I never noticed a case where "move forward" was chosen when there was a wall in front of them, nor "turn" when they were in the middle of a corridor.
- testing prompt (were llms instructed to progress in game, as opposed to just explore — the author said smarter llms were more likely to explore)
- benchmark with humans