Claude Sonnet 4.5 is out. How far will this one get? Will we finally pass Rocket Hideout?
idk i'll check back in four months
this shit was old a year ago who cares
Surely Safari Zone is, by definition, a guaranteed dead-end for this game?
>>722037560who gives a shitall that matters is my MILF sex chatbots get smarter and respond more naturally
>>722037614It gets smarter with every update.
>>722037670You are the lowest tier IQ of AI user. Just wanted you to know that. Thousands of productive implements and yet your life and brain have been reduced to ape instincts. Have a good day
>>722037869pretend i greentexted your post with a soijak
>>722038102based anon owning that faggot
welcome back /v/shame he didn't pick bulbasaur again
>>722037560Its tiem
>>722037869>posting this on a skinner box forum
>>722037654didn't all the other ai's get through the safari zone?claude can't even make it out of rocket hideout LOL
I didn't pay attention to the last run at all. Did it just go the same was as Claude 3 and stall at Celadon?
not gonna watch him do the same shit for the 10th time
>>722038614Pretty sure the other AIs were guided tours while claude is just wandering around on its own
>>722037560This shit is not even the AI playing the game. It's actually """thinking""" on a shitty grid overlay instead. When it actually has to use the AI, it can't even understand what the player sprite is the player.
>>722038820more like, the other AI were able to look at the screen
>>722038727Ya, pretty much. It did make it into Erika’s area of her gym, but he went full retard after defeating every other trainer and somehow never speaking to her.
Zubrats
>>722037560Based Squirtle chad.
>>722038504Opus was significantly dumber than the older Sonnet a lot of times, so this is good.
>>722038820Gemini and GPT had a navigator with pathfinding since they could see the screenThat's why
>jeets are still watching this shit>jeets are still defending it against other AIs that managed to pull it off
...you know retarded infant babies beat this game back in the day? they were the target audience? they didnt have internet and all the guides were full of bullshit?
>>722039843you can't just call everyone you don't like jeets, anon.
>>722039843Difference is Claude will eventually do it without the training wheels those ones had.
I'm liking the new purple/blue coloralso HYDRO THE SQUIRTLE
>>722040386Dev said Claude picked the new colors.
>>722040627it reminds me of the old gbcvery fitting for this
You skim over zubrats again and I'm walking.
>>722040828he will catch one this timetrust the plan
>>722040828Dude is getting confused on route 1, i think we are in for blackout stretategy mk4.5.
>>722037869Yeah, that's disgusting. Ny loli Heather Mason chatbot is much cuter and more engaging than some dried up hagbot.
If he doesn't make progress this time I'm done!
>>722037869good morning sir
Claude seems smarter, this worries me...
>>722041702>read this>check the stream>he's in Oak's lab panicking he can't find Oak
>>722041880kek
>>722041702Opus was really dumb with this game despite being the bigger model. It only made it to celadon as fast as it did because it was more random and stumbled onto the right path more often.
Professor Oak might have gone home
WE ARE SO BACK
going to play crystal while watching this
>>722041702>Claude 5.0 gets smart enough to exploit pokemon red's glitches to give itself access to streamer's computer and spread from there
>>722037560>Claude Sonnet 4.5 is outThanks for letting me know. Time to test it out with my coombots.
>>722042967No problem
this pidgy is fucking him up
RIP
>>722043835>attack missedHydro bros...
FUUUUUUUCK
excellent start
good start
IT'S OVER
PWNED BY A PIDGEY.It's already over isn't it?
Did he ever make it to Erika? I know he was one input away from cutting the tree
>Perfect!
>>722044016based retard
starting early with the blackout stratsi have big hopes for this one
>>722044016F
>>722044016This Claude is a master of his craft
>>722044015He made it to Erika, was standing right in front of her, then left because Opus isn't interested in talking to NPCs.
>>722039843>other AIs that managed to pull it offHas some generic LLM style AI actually beaten the game? Only one I know of is that one where the guy had to help Gemini constantly.
>>722044016Great success!
>>722044016Blackout strategy THIS early? Great Arceus, this Claude is on another level.
>>722044190None of them have actually beaten it. They just gave them full maps and instructions to get past the hard parts so they could say their model beat it. Claude is the only one to even try to do it without assistance.
He's now optimizing the context instead of cleaning it up.
>>722037560Map will need an update.
I just named my rival Claude
CLAUDE GODS WE ARE SO BACKRIP THE PREVIOUS RETARD WHO ALMOST FOUGHT ERICA BUT WAS TOO RETARDED KWAB
>>722040828>>722040992He caught a zubat last time. Problem is it went straight to PC and you know how he doesnt care for computers
he's going to job to the rat
Did they at least fix the thing where Claude has no way of discerning a cuttable tree from a non-walkable tile in the data he's getting fed? His vision is clearly too shit to see see the trees so it's always down to luck if he decides to use cut or not.
>>722037614... didn't this start in March of this year
>>722045380>Level 5>cant escapeThat’s a big rat.
>>722045416Doubtful. The hope is he’s smart enough to discern it now (probably not)
>red againjust pick another fucking game alreadythis stopped being interesting after claude 2 died
>>722045380YAMERO
>>722045551It’s not meant to be interesting retard. The idea is to benchmark his progress on one specific game. Dev streamed it just because he thought others might see it, but he always running the same thing at home for each iteration of claude to see how well each model improves compared to the last one.
holy shit, he's not turning around to get healed?
>>722045719blackout counter 1
>>722045769So much good OC made for these threads
Will Claude ever recapture the magic of Claude 2?
>>722045958never trust RAMmisty lives in the trashed house
>>722045958>6 months later>36 views on a new runYeaaah probably not but I still think it’s interesting to watch each version. Most have moved on
brosClaude has rejected all forms of thinking
>>722045958you never know with these retards
He’s going to die in the suicide forest
thanks for proving that AI has no chance of replacing any jobs worth a damn when it cant even finish a game for five year olds
>>722046148only sailors know how to leave
will he be unable to run a third time in a row?
>>722046336Hopefully he jobs to a weedle
>>722046434soon
>>722045958He needs to bring back the schizo multi-file memory
in the history books, is the benchmark for achieving AGI going to be an AI that can beat pokemon?
rip (again)
>>722046434>>722046469ITS HAPPENING
>This is EXTREMELY bad
We're off to a great start it seems, dying twice before Pewter
>>722046434>>722046469>>722047013LMFAO KWAB
I'm starting to think Claude might be kinda fucking retarded.
i appreciate this one's spirit for insisting on exploring despite his only pokemon being at 9 hp
Is this the most BALLSY claude yet?
IS IT OVER?
>>722047149That’s crazy talk, imagine if SAND heard you saying that
>>722047240It’s kantover.
Anyone else turn up the volume during tense segments like this?
AAAAAAAAAAAAAAAAA
Can he see better yet
Death #2 before the first gym. Can he make it 3?
F
And with that, STING’s bloodline has revenged his neglected ancestor
>>722047498>the Bug Catcher trainer will stay defeated
>>722047539Love this new cope
i feel bad for this squirtle
>>722047654He smokes that HYDRO to numb the pain
>>722047654It's the squirtle's fault. If he cared about winning he wouldn't have let that Ratata damage him so much. This new and improved Claude will make his Pokemon learn that lesson.
>>722037560Does this one have a soul?How 'OnlySailorsCanLeave'-pilled is he?
>>722037869Using AI to avoid women's bullshit is the most high IQ move in the history of mankind
>>722047498Why is he self inserting as the pokemon?
>>722047895this one is suicidal
>>722037560hopefully it wont cheat this time
>>722039843>Gemini it's real AGI SAAR!
Has this been tried with Gemini? It's very smart. Claude seems dumb
>>722049665there's a gemini plays pokemon and a gpt plays pokemonthey got further since the run is handled differentlythey can actually see what's on screen, and they have an active pathfinder to get through complex areas in one series of inputsthey have their own faults but they could overcome rocket's hideout in addition to safari zonebut they're more to be compared with each other than claude
>>722049905also it's worth noting that gpt eventually moved onto playing crystal, which it got through without any real barriersit was just underleveled for the E4 at first but correctly understood it could scrap its way throughthen it did post-game legendary hunting (even though it killed suicune) some more E4 farming for money for ultra balls and then did another run with meganium instead of typhlosion which it didn't seem to finish for whatever reason
>>722050264in terms of finishing i meant beating red, which the typhlosion run did (lanturn 58, typhlosion 62, graveler 54, pidgeot 51, kingler 36) and meganium seemingly did not
>>722047987>let sexual instincts be redirected by AI>fail to reproduce>die out>????>profit
>>722051101NTA but I consider myself unattractive and would not want to bring a child into this world anyway because I know I could not support themThere's ten billion people on the planet, humanity won't go extinct if a few like myself don't do their job.
>>722045958this is a actually a pretty apt example of AI as a wholemassive gains with each new version at the start, tons of exposure and people tuning inthe big gains are gone now. Every new update, while an undeniable improvement, are no longer advancing at such a rapid rate. the exponential gains are gone.less and less training data
>>722037560Imagine industry spending billions of dollars into text autocomplete in steroids and using that to beat pokemon
>>722037779So does my smartphone, but you don't see me bragging about it here.
>>722038102Powerful.
>>722045958We got so little done but nothing was funnier than watching that tardbot lose it.
>>722052554Eventually he just went off the deep end and was lost forever in Mt. Moon
Appears to be stuck thinking it's far enough to reach the end atm
>>722052730We're finally seeing the results of inbreeding.
squirtle is a fucking digger
>>722047987don't mad when Immigrants replaced you
>>722039248>but he went full retard after defeating every other trainer and somehow never speaking to her.hahaha no way, I saw him in there fighting the other bitches and went to bed assuming it was a done deal
This shit is embarrassing. The other major LLMs beat Pokemon in like 2 weeks. This convinced me Claude is complete ass.
What is even the point of this run if he can't see shit?
>>722055859this has been answered like 5 times in this thread alone, they use cheat harnesses and pictures of the screen to let it actually see, that is the distinction
>>722052019What does that even mean? Phones aren't vidya. I fail to see any relevance
>>722056381They all get screenshots. The other ones just additionally get heat maps and tailored instructions to get past some of the stuck points.
>>722055591ESL
Does this Claude have any pokeballs?
It seems more retarded somehow.
>>722058964Yep. I think he has a new “context” system so maybe that’s why
This Claude doesn't "think", he ANALYZES.
>>722059348He analyzes deez nutz lmao
>>722044863VGH fine.....didn't know if he popped into route 22 or not so let me know and I can update it or fix whatever else I fucked up
also RIP clud 4 (I don't actually know if he died in the spin zone but spiritually he never left so whatever)
>>722061693>>722061814SOVL
>>722061814rip spinning retards
>page 10it’s kantover
>>722056458Claude does not. We're not memeing when we say he literally cannot see. He gets an abstract, almost ASCII based map of what's around him. People have posted illustrations before
>>722037560Bump
>>722061693sovl
>>722051796uhm. g-game name?
>>722066235No, he does. The dev's own diagram of the harness as well as interviews confirmed that claude gets screenshots, with coordinates overlaid so he can tell where things are. Then he ALSO gets a shitty little ascii map of the visible screen in his context.People here have made illustrated mockups deforming the screenshots to attempt to simulate claude's retardation, but he does get actual screenshots.
>>722040048>without the training wheelsExcept claude also gets hints and instructions from the dev you retard.
>>722045769Love the aesthetic on this one.
where is claude now
Claude will NEVER pass Rocket Hideout as long as he uses the navigation tool, has a short term memory, and take images only when he's moving.
>>722037560Wait, did it ever finish the game? I stopped following it ages ago.
Since the streamer got rid of VODs, is anybody taking screenshots (and saving them) to compare step milestones?For example, what step current Claude will beat Brock/Misty/SS Anne/Surge compared to the previous version?I want to see how much this version has "improved", and I think so would most people.Really sucks balls that the VODs are gone, since I was the one who posted step comparisons in the previous run.
this retard can't get out of the forest nowit's over
For the previous run, why did Claude switch Leaf to be 1st in the party? Was he training it? I think his level went up. But why would he be leveling Leaf? What was his reason?
Oh, a new run, and he picked a good starter this time, good.
>>722071061to make its cuts more accurate
blackout counter is still on 2?
>>722071109That sounds retarded enough to be believable.
>>722037560Does it want to see you, all of you? It asks, its ministrations reverent, barely above a whisper.
>>722037869do not redeem
Didn't Claude 1 die here?
>>722070652IIRC Highest point is reaching (not battling) Erika, some other models have finished the game, but i don't think they been as open as Claude about what is happening under the hood, and i think the first one that did had a PROMPTER babysitting it, making Claudefags dismiss it.
>>722037560i hope he's still dumb
>>722071972 I miss them
oh god, here we go againclaim your lass
>>722073072green lass
Claude IV's final team for the record. BLAZE never 'zarded... (I'm aware SPIKE had Bubblebeam towards the end and not Water Gun but lol lmao)
>>722073202Will Spike make an appearance again?
>>722073202And here we have Claude IV's final BOX. https://www.youtube.com/watch?v=4xqFVWqM-k4
>>722073308for sure, i think he caught spike in every iteration so far
>>722073321man drowsee had so much meme potential i hope he catches another
And finally, Claude V's humble beginnings.
Favorite Claude IV moments? For me, it was the Route 8 Hypnomassacre. I was fucking dying watching that shitshow. His fixation on Water Gun was also amusing.
>>722073461>claude /v/This is going to be good
>>722073839when he started naming his ratatta or when he started circling the girls in erikas gym like a shark
>>722037610Claude expert here, claude is only as smart as the tools the user provides, this fucking egghead trying to get Claude to map pokemon is basically blinding Claude, like 95% block on what he should be doing. Claude literally can do anything, so Claude in this twitch channel is literally fucking choking on a lack of proper data.
>>722073839When he thought the wall poster text was a team rocket member guarding it. He started counting the amount of rocket members every time he read it and thought they would eventually run out of guards.
Why do you keep posting this streamer?
>>722074335The vision issues alone have absolutely gimped claude in these runs.
He's almost out of the forest.
We made it.
>>722074335The point isn't to make a claude that can beat pokemon but a claude that can play pokemon. Arguably he's still getting too much help with the navigator and forbidding mixing movement and selection in menus.
>>722077261What I don't get is how in the fuck have they not trained these LLMs on basic 2D navigation? They're still at literal retard levels. Actually even a retard could navigate better (I see them on the train sometimes and they get off at their correct stops).
>>722077631Not much use case for it, I guess? It'd take effort to make a dataset for it for not a lot of gain.
>>722052019You literally just did.
>>722077725There are a million uses for it if they want them to be used as agents navigating 2-dimensional layouts on a screen. I guess even billionaire tech wizards can be absolute fucking retards.
>>722076624>NPCs scattered aroundHow did its vision actually get worse compared to 2.0?
>>722078009From the benchmemes this one was optimized for agentic behaviour. It backslid in some areas as a consequence (though amusingly image recognition did improve)
>>722078139Funny how these LLMs are supposed to be getting closer to AGI but really they're just getting more specialized for agentic use, and every other area either suffers or remains stagnant but never improves anymore.
This stream is all but dead. He's fighting Brock now and barely anyone is watching.
>>722080676Im in bed waiting for updates
1ST BADGE GET!2380 STEPS!
brock defeated at 2382 stepshow does clod the 5th stack up compared to his ancestors
BACK STATUS???
>>7220810243500 I believe
Iconic picture
>>722081118We're
>>722081118SO
>>722080676It's the fifth run so a lot of people just got bored and the few that remain don't want to get invested too early. Can't blame them because without improvements to the navigation tools this Clod will most likely speedrun the game until Spin Hell and get stuck there for months even faster than Cl4ude.
Guess I'll go ahead and update this, even if he'll be in Mt. Goon soon
>>722037560Why not play gold or silver to mix it up? I'm tired of seeing the same failed run over and over.
He hasn't caught anything yet so the chances of filling a spot with Zubrat early is high
What do (You) want claude to catch this time?
>>7220824925 clefairys
>>722082487Fingers crossed.
>>722082569will SHE retvrn.....?
I am not crazy! I know the SS Anne departed. I knew I needed to go heal Sand. One after Sprou. As if I could ever make such a mistake. Never. Never! I just - I just couldn't prove it. The SS departed, he got that idiot with the red hat to lie for him. You think this is something? You think this is bad? This? This chicanery? He's done worse. That market! Are you telling me that a decoration just happens to be there? No! *He* orchestrated it! The captain! He *defecated* on *Sand*! And I saved him! And I shouldn't have. I took him back to the Pokécenter! What was I *thinking*? He'll never change. He'll *never* change! Ever since I caught him on route 2, *always* the same! Couldn't keep himself from missing! "But not our Sand! Couldn't be precious *Sand*!" Getting burnt to a crisp! And *HE* gets to be healed? What a sick joke! I should've blackouted him when I had the chance
>>722083359
>>722047240It would make sense for the grimer girl to be into ugly bastards.
>>722084304hot
>>722074335>can do anything>can't handle a miniscule amount of self-agency
How's he doing compared to Claude 4?I remember the previous run was fucking zooming through until Vermillion
>>722085787A bit slower, also less Pokémon.
>>722085787worse, this retard died twice before reaching brock lole
>>722037869Why do trannies get so angry when people jerk off to normal shit instead of degenerate slop?
>>722037560The devs were so embarrassed by the last attempt that they secretly programmed instructions to beat common games. Smoke and mirrors.
remember when claude could only move one step at a time?
>>722086927Better times.REMOVE NAVIGATORREMOVE NAVIGATOR
>>722086625>>722086669Have we reached the limits of LLM ai? Chat GPT's newest model was also worse wasn't it?
>>722086797He lost to a Pidgey
>>722051270See the problem is that only white people think like this.
>>722087002Yeah apparently GPT5 was a step back. Bound to happen eventually.
>>722087042Forget the bird he lost to fucking weedle
This fuckin thing takes 15+ to read the screen AI is so retarded
SQUIRTLE IS ABOUT TO EVOLVE
So what's the endgame here? AI beats the game = AGI confirmed?
>>722088174AI beats the game = cheating confirmed, do it again with less harness.
>>722088256If he was gonna cheat, he would have done it three or four Claude's ago.
>>722088327Yes, no claude has beaten the game therefore no claude has cheated. When(if) claude ever beats the game then it was clearly cheated for [reason]
>>722087002My own hypothesis / understanding is that we ran out of training compute. Won't be able to increase general intelligence before Stargate is operational in 2-3 years. Until then it'll just be improvements at benchmark tasks and agency.
Somebody keep track of all the major milestones, and compare the number of steps needed to complete the task.>beating Brock>exiting Mt. Moon>beating Misty>leaving SS Anne>beating Surge>entering Route 10>exiting Rock TunnelThere used to be a graph comparing Claude models and the steps needed to complete these tasks. But it hasn't been updated since 2 models ago.Getting harder to do because no VODs to look up the steps.
