>It's still running>There's still people periodically checking this shit
Clade will never pass Celadon
Geminichads stay winning
Why does this thing suck so much? I've seen AGI learn how to walk and how to break trackmania records. You'd think they would be able to play pokemon. I imagine this works differently than the ones I'm talking about.
>>726010882they intentionally gave claude as few tools as possible. it's why the other LLM's can beat it (with human intervention at that) and not this one.
>>726010882correct.every other thing has had to have custom shit programmed into it in order to better interpret the world.gemini had full on cheatsheets of instructions to solve mazes step by step put in.the idea with claude is one dev put together something barebones that mostly relies on its own puzzle solving and limited image recognition, with some raw data like its pokemon, their state, and where unwalkable tiles are.they actually had to restrict its ability to take notes early on because it was such a spaz it managed to fill its context (which uses RAM) to the brim with abandoned files of notes.
>>726010882No you didn't, there's no AGI.It sucks because they're trying to have it succeed by itself, without guiding it.
>>726011034It sucks because it's a fucking generic LLM and trackmania used an RL model trained on... trackmania. Niggas itt are retarded as fuck
It's a better way to spend your time than playing Skyrim.
>>726010882>Google and OpenAI went above and beyond to make the game as unfairly easy as possible for their own models to ensure they would finish it>Anthropic got cocky and did the absolute bare minimum so it backfired There was no middle ground.
>>726010616Damn it's been awhile, last time I checked in he was in the never ending diglet cave into mt moon loop. Think they lobotomized him several times but he never got out of the loop.
>>726010616Still perpetually on my second monitorStill an undefeated hopechad
>>726010616Wish they tried a different game already.Maybe he will be able to play games with less "navigational challenges" and more straight logic/math, like card games.Claude plays YGO GX Tag Force when?
>>726011482Spitechads won during Opus 4 so you can go fuck yourself.
>>726011243Ah ok thanks. I don't know much about this. I was confusing RL models with LLMs and using AGI as a blanket term for both.
How fat did Sand end up getting?
>>726011019>it was such a spaz it managed to fill its context (which uses RAM) to the brim with abandoned files of notes.Holy fuck Claude is literally me
>>726011548>I was confusing RL models with LLMsCongrats, now you know more about AI than 99% of the world.
>>726010616>>726010882Its not AGI, its an LLM. its literally just Claude as an instance being fact checked by another instance of ClaudeAlso the total memory it has is extremely small and literally runs on a token allowance as its the pet project of one of the devs. Clod cant even SEE most tiles and has a hard time telling where he can even walk. >>726010669 is right it might be genuinely impossible to leave Celadon just because of how poorly designed it continues to be
Wait this shit is still going? I haven't looked at it in like two months.
>>726011590like 65 kilos (levels) or something
>>726011590>>726011748Sand was sexo too bad she blimped outAs least we have the eternal goonslop
>>726010616>walkthorugh information what does it mean by this? it's not playing blind?
>>726012680Ever talked with Claude? If you ask him information on how to beat a game, he will likely reference walkthroughs.That's the very same chatbot playing the game, but there is a huge difference between calculating the correct answer to a question and actually figuring out the inputs from his limited vision.
by the way someone already figured out how make claude retain actual important context and navigation but we are stuck bruteforcing llm brainfarts
>>726012970>Ever talked with Claude? >If you ask him information on how to beat a gameThat shit is too expensive to be asking how to beat a goddamn game
>>726013205You can just ask Sonnet 4.5, the one playing the game right now for free.
>>726010616>There's still people periodically checking this shitThe summation of my "checking" is just seeing the stream thumbnail on my Twitch following page as I open it up to watch someone else. And it's always Celadon City outskirts. Sonnet 4.5 hasn't even gotten to the Rocket Hideout. The prior Opus instance was smarter.
>>726013270Hmmm... I guess.But I use all of that with SillyTavern so there's no access to Claude from that anymore.
Even if it's over now, I'll never forget Claude's schizo meltdown followed by him solving the SS Anne and speedrunning Surge, really something that has to be seen to be believed
>>726013383People always focused on the "only sailors know how to leave" part of this hallucination, but for me it was Claude saying "we don't go outside when..."When WHAT, Claude??? Why don't we go outside when what happens???
>>726010972>(with human intervention at that)So they didn't beat it.
>>726013383Lot of funny OC was made. The "only sailors know how to leave" schizo moment was gold.
>>726010616Why don't someone just make a game with dynamic NPCs using LLMs to think instead? No baked in schedules for each NPC, they just ramble to themselves if they need to do something and will do it ingame.
>>726014672Because not everyone has a machine running 5 3090's.
>>726014707I'd imagine you would be using an API like Claude or ChatGPT, not running it right on your PC.
>>726014057Correct
>>726014791>I'd imagine you would be using an API like Claude or ChatGPTNigger, I'm not paying for shit. t even pay for the game itself it it existed.
Claude play Master Duel thenLet see how well its play agains that autist magnet
>>726016807I find it odd to me that the franchise is still called "Yu-Gi-Oh!" when only the first series had a character named Yugi in it, and the series then went the way of alternate universes whose only commonality is the card game Duel Monsters.
>>726010882It fucking sucks because the second reset was absolutely fucking soulless
>this thing has been draining megawatt hours of energy every day for months for zero benefit of any kind to anyoneIn a better world the guy running it would be prosecuted
>>726010616it's kind of funny that hitmonchan/lee are named after chinese martial artists who were famous in america instead of the actual boxer and kickboxer they're named for in japan.
>>726017624He works for Dario so that filthy jew lets him do it for free
>>726016807Master duel would actually be easier because there are fewer valid moves. It's like how chess isn't hard for algorithms much less LLMs.
How about PvZ?This seem easy enough to play, yet require actually strategy to win
>>726018557highest LLM chess ELO is like 850 tf are you talking about. the chess models are specialized
>>726013431 I swear that line lives rent-free in my head. Claude just stops mid-thought like he's about to reveal some deep eldritch truth he wasn't supposed to access. "We don't go outside when..." WHAT? When the moon's inverted? When the pokeflute goes silent? When Bill becomes many and one? Absolute creepypasta tier and it was just casually tossed in mid-reset loop lmao.
>>726018684delay between thinking and acting would likely fuck it over
Why not make AI play something like Disco Elysium
>>726018919Maybe using PvZ Fusion mod that modified to slow down enough?
