World models don't need to be taught again. All the current top AIs are capable of solving Pokémon just from what they've read on the internet or their general knowledge.
I think it is possible but seems quite difficult to me from this experiment. The hard part is that Pokemon has a lot of different goals, and has many strategies to complete it. You would need a way to encode these intermediary goals in some way temporally that the model needs to learn to complete in order. For example, you need to collect all eight gym badges before you can challenge the Elite Four to win.
You could compose multiple network, one with the goal of learning how to interact with the world through the screenshots. This would probably require a lot of frames of the game from many diverse locations and scenarios--as the network would need to learn how battles works, items (though technically not strictly necessary), and general dialogue and menu interactions.
Then other networks could then try to learn to encode different intermediary goals trained on a bunch of noisy runs that complete that goal. Collecting this data seems tedious and difficult though and is a whole project in itself.
So my intuition is yes, but I don't think simply making my current model bigger would get us there. I see you have experience with world models--what would you try? ;)
That makes sense, thank you! Although it's telling that, with language understanding, a human player is able to discover all the goals for themself, rather than needing to have them specified and shaped up-front. Seems like we still have a long way to go on that front :)
And in terms of what I'd try, I think it's a really hard question haha! I'm interested in the idea of hierarchical world models, where you have something doing prediction on a long-horizon, abstract task level (if I beat this gym I can fight the next one) as well as a short horizon model predicting over individual inputs moment-to-moment. I believe LeCun was actually attached to a hierarchical JEPA paper for robotic control - but it's still very early days.
Art is a conversation between two people, that is a timeless communication between human.
An artwork to realize... to surf within to learn from, to listen to, to realize someone's images, someone's ideas, someone's worries, someone's messages... someone...
Art is someone's work of life, their experience, their... love...
There's human time in art... in a book, article, painting, soundscape, film, video-game... there's alive time of other human who reified their life experience, views, ideas... in an artwork and passed it to the outer world... to be inscribed in the infinitely magnificent history...
This...? This is just, sorry, deadly awful, dreadful, dark, and dear pain in heart... This is also disrespectful towards the Artists, Developers, people who worked on the Universe and this exact artwork...
This is just a mathematical model playing a game from the 80s, it's not even trying to mimick artwork. How can you say it's not art to teach a dog how to play chess? Is it less of an art then to teach a machine to play Pokémon just because the machine isn't alive?
Dare we have two AI-plays-Pokemon threads at the same time?
Show HN: Jev Plays Pokémon Red - https://news.ycombinator.com/item?id=49845172
I think seeing that thread on the front page gave me the courage to post my own post!
YOLO.
World models don't need to be taught again. All the current top AIs are capable of solving Pokémon just from what they've read on the internet or their general knowledge.
At the end of rabbit hole you meet The Bitter Lesson like everyone else.
Author here, feel free to ask me any questions. Though I am no expert in this, just someone trying to learn and have some fun doing it.
Curious on whether you think this JEPA-style approach could scale to a full game completions?
For reference, standard PPO has been able to beat the game end-to-end with a relatively small network https://drubinstein.github.io/pokerl/
I think it is possible but seems quite difficult to me from this experiment. The hard part is that Pokemon has a lot of different goals, and has many strategies to complete it. You would need a way to encode these intermediary goals in some way temporally that the model needs to learn to complete in order. For example, you need to collect all eight gym badges before you can challenge the Elite Four to win.
You could compose multiple network, one with the goal of learning how to interact with the world through the screenshots. This would probably require a lot of frames of the game from many diverse locations and scenarios--as the network would need to learn how battles works, items (though technically not strictly necessary), and general dialogue and menu interactions.
Then other networks could then try to learn to encode different intermediary goals trained on a bunch of noisy runs that complete that goal. Collecting this data seems tedious and difficult though and is a whole project in itself.
So my intuition is yes, but I don't think simply making my current model bigger would get us there. I see you have experience with world models--what would you try? ;)
That makes sense, thank you! Although it's telling that, with language understanding, a human player is able to discover all the goals for themself, rather than needing to have them specified and shaped up-front. Seems like we still have a long way to go on that front :)
And in terms of what I'd try, I think it's a really hard question haha! I'm interested in the idea of hierarchical world models, where you have something doing prediction on a long-horizon, abstract task level (if I beat this gym I can fight the next one) as well as a short horizon model predicting over individual inputs moment-to-moment. I believe LeCun was actually attached to a hierarchical JEPA paper for robotic control - but it's still very early days.
Why is this a dupe of the other thread?
This isn't, they are two independent projects making two independent models play Pokémon... independently
This is so dear sorrowful...
Art is a conversation between two people, that is a timeless communication between human.
An artwork to realize... to surf within to learn from, to listen to, to realize someone's images, someone's ideas, someone's worries, someone's messages... someone...
Art is someone's work of life, their experience, their... love...
There's human time in art... in a book, article, painting, soundscape, film, video-game... there's alive time of other human who reified their life experience, views, ideas... in an artwork and passed it to the outer world... to be inscribed in the infinitely magnificent history...
This...? This is just, sorry, deadly awful, dreadful, dark, and dear pain in heart... This is also disrespectful towards the Artists, Developers, people who worked on the Universe and this exact artwork...
This is utterly sorrowful...
This is just a mathematical model playing a game from the 80s, it's not even trying to mimick artwork. How can you say it's not art to teach a dog how to play chess? Is it less of an art then to teach a machine to play Pokémon just because the machine isn't alive?
Small correction but it came out in Japan in 1996.
Wopsie... I had it when I was a kid so late 90s and I remembered the Gameboy itself going way back in the 80s so one assumes
Why are you on hackernews, jesus
Unfortunately, the developers of PGo! knowingly utilize the play of innocents' to train military SI.
inb4 AI is lost to SI. Irony zone /s