Topic.
Context: Every few months I go back to Blue Prince. It isnāt a big game, so I havenāt uninstalled it from my Xbox. I app bought it on Steam, to play on my Mac, but it plays better on Xbox, so thatās where I play.
Minor spoilers in the next paragraph.
Iāve never gotten to Room 46. Iāve gotten to the Antechamber. Iāve been underground, I basically understand how to beat the game. I also understand that Room 46 just beats the tutorial; thereās way more game past it. I have three of the letters. I have the grove unlocked, the outside shed the garage unlocks, and I have the gem cave. I figured out the puzzle of the two drawings in each room. I know what that message is.
Still, Iāve only failed and I think I have around 50 runs?
Iāve asked Google and apparently nobodyās tried this? I keep seeing a YouTube video with the (bait?) title that itās not RNG.
I havenāt tried this, but I keep thinking, if I ask a chatbot to play the game, it tells me which way to go, I tell it the three choices, it tells me which to pick and where to go, I think that would work. But is the chatbot guaranteed to win? If not, can the game generate runs that simply canāt be won? Or is there enough RNG that both can be false?
I get that a lot of people donāt like AI. Thatās why I checked to see if it had been done first. If anyone had done it or tried it, I couldnāt find it. Iām really more interested if it can or canāt, or if one canāt but another can, than actually doing it. Most games canāt be played by having a chatbot walk you through it. Those that can ā point and click adventures ā can similarly be beaten by simply following written directions. They donāt have RNG. Something like Life is Strange, for example. You can get all the collectibles, and achieve the optimal outcome, just by following a guide. But, Blue Prince is built differently.
Could a bot (Not LLM) be made to beat it? Sure. It wouldnāt be beating the gameās RNG any better than a good Blue Prince player though. Bots following decision trees based on probabilities is nothing new.
If youāre going to take out the joy of doing the puzzles yourself, you may as well watch a streamer play it. Iād recommend LordMinion777ās play through, he really enjoys the game for what it is.
If youāre only touching the game every few months itās not a surprise you havenāt gotten to Room 46 yet. You remember the basics everytime you go in, but you have to relearn the patterns and interactions so youāre just putting yourself in a failure loop.
lil indigestion also has a very comprehensive playthrough.
Yes, if I just wanted to see the ending, that is exactly what I would do.
I remember more than the basics, but, good point.
Sometimes I feel like Iām not smart enough for Blue Prince. Thatās fine. Puzzles made by smart or smarter people can stump me. That doesnāt bother me, and I still love the game for what it is.
LLMs do not have a world model. They are what is called ātheory free inferenceā. Due to this, they canāt actually even play chess. There is no actual awareness of context (which is part of the linguistic trickery around its features, see also āhallucinationsā). It can, however, provide answers to questions about the level, even though its answers will be based on probability occurrence and not correctness or understanding.
Humans do not have a world model. They are what is called āanimalsā. Due to this, they canāt actually even play chessā¦
The only way an LLM would Beat the game is if itās trained on it over and over and over so itād understand the principal and strategy. However, LLMs arenāt built for that, for that youād be a statistical or probability model AI.
LLMs arenāt built for that, for that youād be a statistical or probability model AI.
Pretty much all current gen LLMs have that capability. Theyāve gone a long way past being just language models.
I figured if it could parse a wiki and all thatās been said about the game, that it might be able to forge a winning path.
Look, I know the go-to lately for a lot of people is ājust ask aiā.
My problem with that, is, itās like cheating on a test. Taking the easy route. Without the critical thinking skills, without the development of figuring out how things work, weāll lose the ability to think differently!
Whatās the danger or risk with that? We become more docile, controllable, herded like cattle or sheep. Taken advantage of by the ultra wealthy because we give up our ability to think and act differently.
There have been multiple studies of this being done that show this is the way weāre going. I also donāt use AI because of the ethical issues behind the technical inefficiencies (very power hungry, race for ābestā) and skip the āshould we be doing this on the first placeā.
I get that. Thereās a whole debate worth having around AI in general, and there are comms dedicated to it. Even people spamming their point of view in every other comm hoping to change mindsā¦
Iām just wondering if it can without trying to touch if we should. There are smarter people than me debating that in other places. While preaching to me might feel like an easy win, it wonāt really address the problem, and I wonāt not seek an answer.
I donāt like AI because itās an inefficient use of resources, but AI used to make something that people benefit from changes the equation, though itās subjective how much it needs to change it to matter.
https://www.youtube.com/watch?v=Fa5el2uhP_8&t=92 LLMs can play Portal. So Blue Prince maybe could be done the same way. It is harder than Portal though.
āPlaysā. Only a few minutes into the part 1 video it preempts the tutorial jump instruction. Before any indication of the upcoming jump instruction it both shows knowledge of it in the reasoning and actually jumps.
Maybe because Portal 2 is an old game probably well described in the training data. Might be interesting to see how well it works on a new game.




