Why is AI incapable of beating even the easiest of video games?

>>717387447 (OP)

Trained on journalists, or women.

>>717387447 (OP)
It's a LLM. Typical inference means it doesn't really learn as it goes, so it will never understand dynamic relationships and will just go in circles.

What kind of retard thought this would work

Oh shit
Is there a new run?

>>717387447 (OP)
He picked Charmander this time? Cool.

>Why is a machine that runs on Markov's chains incapable of performing complex actions that require actual sentience
AI's not gonna destroy civilization, its massive financial crash will

>BU ZZ
I missed watching this little retard

Two reasons.
1. Most big models nowadays are tuned with reinforcement learning to be as overconfident as possible as it makes them look smarter. Claude never tries to reconsider dumb shit like the captain hallucination as it was trained to never doubt itself.
2. LLMs are token pickers, and eventually as the models go on they fall into loops where certain tokens become so common in context that they get picked every single time. This is what leads to loops, and no LLM has even gotten close to solving this so far.

>>717387447 (OP)
Because AI cannot think because it's not actually artificial intelligence like VR isn't virtual reality.

Next question.

>>717388451
>as it was trained to never doubt itself.
Descartes would find that funny.

>>717387447 (OP)
google gemini already beat it tho

He made it to pewter easily this time

>>717387447 (OP)
Word processors are not AI

>>717388249
I sure am glad that this technology will be used to monitor user IDs across the internet :^)

>>717388195
>>717388205
Yeah, started today since Claude updated to Opus 4.1.

>>717389397
Has there been any kino this run?

>>717389505
He caught a Weedle and named it BU ZZ.

Did Claude win any of his prior runs? When I stopped following it was still at 0.

>>717389584
Nah, gets stuck in the rocket hideout

>>717387447 (OP)
AI can beat your mom

here's to 5 more months, maybe 5 more years of Claude and his adventures in Kanto
https://www.youtube.com/watch?v=-z1NR9IQICo

>>717387447 (OP)
>Why is AI incapable of beating even the easiest of video games?
But AI can do fine at plenty of very easy video games. What this program has actually highlighted is that Pokemon is more sophisticated than people give it credit for. Sure, maybe you don't see it as such, but it's not just a linear tunnel with a mandatory puzzle at the end.

Here's a true nuke for you: Pokemon Red and Blue are more complex than Bioshock: Infinite "a smart game for smart gamers". This is a game that involves walking down corridors until you need to interact with something.

>>717389659
Damn. Poor guy, at least he prompted a lot of smiles on his journey.

>>717387993
This. So I didn't understand what the point of this is.

>>717387447 (OP)
Because they're not AI for starters.

>>717387447 (OP)
Because it's a giant grift/meme.

>>717389584
Nope.

1 ended in Cerulean City. Claude blacked out against the Nugget Bridge rival battle and because he returned to the Pokemon Center thought he cleared Route 24, 25, and got the SS Ticket from Bill. He got stuck in a loop trying to escape Cerulean City through the trashed house and the run got reset.

2 ended in a wide loop around western Kanto. This is the one most people remember, he cleared the first three badges, but then got caught in a loop trying to get Flash for Rock Tunnel, going back to Viridian City and back through Mt. Moon again, and again, and again. Eventually he got Flash, but never found the path to Route 9 or the cue to cut the tree there. This would be a premonition for a reoccurring problem.

3 got the first three badges lickety-split, but then got caught in Lt. Surge's front yard because he couldn't figure out how to cut his way back out. Like 2, he tried to get Flash for Rock Tunnel, but failed. Instead, he actually found the path to Route 9 and cut his way through it, and got to Rock Tunnel without it. Then the API crashed and the dev, on vacation, didn't catch it for like a week. When he returned and reset the API Claude had already cleared Rock Tunnel, passed through Lavender Town and Route 7/8 to Celadon City. This Claude then spent several months in the Rocket Hideout getting confused and dizzy by the conveyer belt puzzles, with intermissions of escaping and trying to find the Celadon City gym, protected by his eternal nemesis, cuttable trees. There's more cuttable trees inside the gym too, and at least one is required to reach Erika, so he was totally boned in this avenue.

So now we have 4, Opus 4.1. The first one with Charmander.

>>717390112
Did that Gemini bot cheat or something

>>717387447 (OP)
Oh boy I can't wait to get back to Celadon. Just got to get through Mt Moon again...

>>717390151
Effectively. Many of the other AI bots' devs favored entertainment instead of an actual test, so the Gemini dev intervened a LOT in the Gemini run whenever the bot got stuck. The dev for Claude is extremely hands-off, likely to the point of full neglect, but it also lets Claude just run on his own without a human playing with divine intervention.

>>717390112
Thanks for the very informative summary. I dropped off in 2, while waiting for him to find Route 9. I'm actually surprised he never did, even though finding trees was a perpetual issue.
>When he returned and reset the API Claude had already cleared Rock Tunnel...
So Claude made a ton of progress without any API access? Not the Claude API then, but access to some of his tools? I don't understand how he accomplished anything here.

>>717387447 (OP)
Use Gemini instead.

>>717387447 (OP)
We love Sand here.
We hate Puff here.

>>717389896
To demonstrate to the average person the limitations of a LLM

is Claude a chad or an incel?

>>717390387
Sorry, I was a tiny bit off. It was OBS that crashed, not the API. So Claude still made progress, but nobody could see it and thought the experiment was over. People suspected the dev might've pushed Claude through Rock Tunnel during the downtime given Claude was having trouble with it before the crash, but I doubt it, given he got stuck in the Rocket Hideout for two months.

>>717387447 (OP)
>>>/vp/

>>717390387
Other anon was wrong. The API worked fine but OBS crashed for 3 days while the dev was on vacation. The bot was still running and accomplishing things, but nobody could see it.

>>717390484
Yeah right. Puff was larger than life from day one. Sand had to work and steal to get any size worth notice.

REMEMBER
CLAUDE LIKES TO RAPE
HE LOVES TO RAPE
REMEMBER

OH SHIT, NEW RUN?

>>717390112
Why the fuck do they not try FR/LG where the highet fidelity sprites would help its shit vision?

>>717390807
old good new bad

>>717387447 (OP)
Ai is just Google with an extra step.

what's gonna happen in 5 years when companies will to have been hyping up this ai shit for a decade and there has been no tangible gain to productivity from it?

>>717390924
I actually have the secret formula for AGI. But i will not tell you guys, nyahahahahaha!

I checked it the other day when i randomly remembered and was surprised that it was still running at all and there were like 20 people in the twitch chat, probably all just people that have it running 24/7

methinks Anthropic is lying about the exorbitant cost of claude 4 opus if they can afford to leave this shit running 24/7 with not only a very negative return on resources, but probably a negative return on publicity as well.

>>717390924
They're already trying to suppress it lol

>>717391079
I think it makes programmers smarter rather than anything. Claude is a thinking thing, in the long term this will help.

>>717390924
You sound like an incel. I invested 100k on Nvidia.

>>717387447 (OP)
This is still going? Last I heard it got stuck in Mt. Moon and was never able to get out

>>717387447 (OP)
the AI doesn't have the "will" to do anything. so it won't complete the game. you have to convert artificial checkpoints in the game that if achieved will reward the AI, but the problem is that the AI isn't even aware of these rewards, so how would it come across them except accidentally

why are you freaks still watching this crap?

>>717391165
>Bringing up sex out of nowhere
kek
insecure much?

>>717391079
Sounds pretty good that they're properly taking their time now. Imagine if the tea devs spend 19% more time on their project instead of rushing it out the door.
Also it's all dumb boomers that hate learning new stuff they're testing this on.
The real test is to give a two guys with no coding knowledge in order to really see its worth. One has no AI to help create a program, one gets to use the AI to help him. You'll probably see the difference in speed to workflow and completion by a very large factor then. A number of years maybe.

>>717391327
you have to tell the AI how to beat the game because it doesn't even know what a game is like we know what a game is. for example, you have to tell it that it needs to beat the gym leader, but how the fuck is it supposed to figure what or where a gym leader is? you have to tell it to talk to NPCs, explore, acquire pokemans, etc. but you have to explain to it what those are in the first place.

>cheated shit again
What's the point? He's had specific instructions for the sake of content for a few versions by now. I thought the fun of this shit was to see him struggle on his own

>blocks AGI from happening

>>717387447 (OP)
All AI models, including generative/LLMs are just complex tangles of layered algorithms built in a black box with the directive of using probability to pick an output that will get a passing grade from external evaluation. All the data it gets trained on is to allow it to refine its ability to weigh the odds against a number of variables and provide an output that gives it a pass.

A pass/fail probability machine does not experience reality. It experiences numbers and, based on what output it has produced in response has earned it a passing grade, it determines an output that will likely do it again. It has no such model to follow in these scenarios. It is developing the successful output raw, with no cheat sheet. So it will take it years, or decades. If you fed it inputs from successful playthroughs, you would have the equivalent of Gen AI and the pwetty pictures it makes for you. A copy of successful input/output scenarios to pattern off of.

>>717391340
we have nothing better to do

>>717387447 (OP)
it is adavancing technology in just many year there was no game ai but now there is game ai tomorrow is very bright hire many ai workers to build future!!! thank you

>>717391486
>Being a schizo again
What's the point?

>>717387993
The fact that these LLMs cannot learn or make complex connections is the exact reason why they're completely overhyped marketing bullshit.

So is there any actual unironic >t. knower ITT who can give it to me straight
Is AGI gonna happen in a couple years' time or not...? Are we on a fundamentally mistaken path for that? Is quantum computing not gonna accelerate that drastically though

CAN FIND STUFF ON THE GROUND WOW

>>717391487
>a 6 year old child could get through this maze in 20 minutes
>it took Clod 3 months to not even make it through once

>>717391079
I am not a programmer or anything, but I did use AI to make a script to automate a chunk of my work. It made things a lot faster for me. I know no programming so it helped quite a lot.

>>717391760
I believe it will depends on how fortune goes on the front leaders of AI.

I think, if we get lucky, we will have AGI in like 5 years. I know it sounds ridiculous, but you never know with current technological progess.
Realistically, its probably gonna take 15 or 12 years. However people probably not realize just how close we are of making a small, intelligent thing that has more capabilities and energy than the average human ever could hope to muster.

>>717391462
>Theyre spending more time so its good
Its 19% more time fixing obvious errors. The end product does not get any better. Infact, its arguably worse because more mental resource is put into fixing more shit, and mental resource is finite. If Tea used AI to code it would have either been worse or takenlonger, it wouldnt magically have gotten better because people fixed obvious bugs caused by the AI itself.
>Also it's all dumb boomers that hate learning new stuff they're testing this on.
Lmao no it isnt dude, stop making excuses for this shit.
>The real test is to give a two guys with no coding knowledge in order to really see its worth. One has no AI to help create a program, one gets to use the AI to help him. You'll probably see the difference in speed to workflow
The guy with zero knowledge that uses AI will make something thats fucking garbage because he doesnt know how to code review and the AI confidently lies and/or shits the bed at anything remotely complex. The non AI assisted person will take longer but almost certainly make a better product because he understands the shit he writes. You dont seem to understand that most projects arent a one-off thing and that they get built upon. If the foundation of the code is shit, it becomes a nightmare to deal with down the line.

Holy shit I'm home again. I don't even know if it's worth uploaded all of the previous run's memes since I imagine everyone has them already.
Does anyone have the final moments of the previous run?

>>717391760
People have been predicting AGI "in 20 years" since I was in school in the early 90s. It's not going to happen, especially with current tech that hallucinates every few thousand tokens no matter what.

>>717391976
the hallucinaiton thing is pretty damn bad, but i think it's also because they train it to avoid saying I don't know

>>717391976
We literally made machines capable of having sex with you with zero human input. To say that things are the same as before is blinding oneself to the implications and pieces of evidence.

>>717391760
>Is AGI gonna happen in a couple years' time or not...?
No, not by an LLM. Like not even close.
>Are we on a fundamentally mistaken path for that?
LLMs arent going to achieve AGI. It is just confidently hallucinating shit. It doesnt "know" what the fuck its doing. Look no further then it trying to play pokemon, it doesnt actually understand whats fucking happening and relies on context that gets wiped very quickly.
>Is quantum computing not gonna accelerate that drastically though
Quantum computing would, in theory, make a huge improvement. Quantum computing is also a meme, and has been for the past decade. I am less optimistic about that than LLM shit, and Im not really hyped about LLMs at all.

AI is only really good for quick general knowledge, but the fact that you need to fact check it constantly makes even that dubious. Its best application is in medical. Outside of low quality cooms and shitposting, anyway.

>>717390807
The classics are more beloved and nostalgic leading to the potential for more people to tune in over FRLG. There's also the fact that there's less for the thing to consider with how much more rudimentary the original games are by comparison to gen 3.

GO CLAUDE

>>717390924
That's when they sell it off to their buddy's companies who hike prices to pay it off

>>717390785
hey fuck you, jigglypuff

>goal is to train spike for 5 levels off of just wild pokemon
please course correct

Does that mean we're back to Zubrat content making hours?

>>717393140
>All to reach Lv.12 so Spike can learn Double Kick
>that doesn't exist in the version Claude's playing

>>717390924
bro I typed fucking "will to have been" how did nobody take the bait

What Claude Opus 4.1 Actually Improves
Enhanced Agentic Performance: The update focuses on improving Claude's ability to handle complex, multi-step tasks and agent workflows. This could potentially help with sustained navigation challenges, but doesn't directly address the core visual processing limitations.
Improved Coding Capabilities: The update emphasizes "real-world coding" improvements, which suggests better software development assistance but doesn't indicate enhanced visual interpretation of pixelated graphics.
General Reasoning Upgrades: While the update mentions reasoning improvements, these appear to be incremental rather than the fundamental architectural changes needed for spatial navigation.
Issues That Remain Unresolved
Visual Processing Limitations: There's no indication that Claude Opus 4.1 addresses the core challenge of interpreting pixelated Game Boy graphics or understanding conveyor belt mechanics. The visual processing improvements we discussed - such as object-centric visual grounding or specialized convolutional components for grid-like data - don't appear to be part of this update.
Memory Management Constraints: While Claude 4 models introduced some memory improvements through "memory files" when working with local file access, this doesn't solve the context window and spatial memory issues that plague maze navigation. The hierarchical spatial memory systems we discussed aren't mentioned in the update.
Spatial Reasoning Gaps: The update doesn't mention improvements to spatial intelligence or navigation capabilities. The fundamental challenge of maintaining 3D spatial relationships across multiple basement levels remains unaddressed.

>>717391079
that study was done by an anti-ai organization using open source github repos that only consisted of over a million lines and only used devs who were previously the main coders for each repo (so incredibly knowledgeable about their own set up) it was also using models that were out around the beginning of the year which didn't have anywhere near the amount of context to handle such large repos. nor did they utilize any kind of RAG system or anything clever like that. but I know a lot of midwits just like to hate on AI. if you look at microsoft's latest earning talk they give specific numbers on how much AI has saved their finances this last quarter

>>717387447 (OP)
We're at the level of mass-information AI, not human-learning AI. They are very different concepts.

I stopped thinking about this after 4chan went down. Good to be back bros

>>717393326
nobody actually reads

>new run

Last time I checked in, Claude was still stuck in Celadon with 220k steps. He never beat Erica and never got through the spin maze (except that 1 time) right?

Assuming his harness is the same, I'll check in when he reaches Cerulean, where he'll be let off his harness. Because I'm assuming he's just following everything up to that point.

>>717387447 (OP)
Because games require above all else understanding, which AI is fundamentally incapable of.

>>717393326
if you were saying something back about ai you would have gotten at least ten replies

but you were saying something bad about ai, so you could have typed 2000 characters of literal gibberish with the string ai and bad in it and it would have been engaged with seriously as if there were nothing wrong
>fhvsbndfkjvskabnhkefbksjjvkbsakjnjnAIhfvbjskbnejfnkjsnjfnjkbadjmcksmleknfklnsklnvklsnk
>yes i quite agree i've been saying the same thing about ai, very well put. Im not sure why everyone cant see what we see AI is clearly the most useless thing in the world sigh it really sucks being smarter than literally everyone else my friend but its our cross to bear

>>717391541
An easier way of saying it is they're just glorified pattern matchers.
IE: highly advanced mad libs.

>>717391830
I thought he made it through once but then went back on the spinners and got lost again?

>>717387993
This.
You need a specific AI for every different part of the game.

Poor Pikachu
He's going to get steamrolled

I want a Bolt rerun.

Claude seems to think faster. Was 4.0 this fast? I skipped most of that run

CATCH IT YOU FAGGOT

>>717393557
anti ai shitters btfo

>>717391760
>Is AGI gonna happen
With the current methods? Lol no
>In a couple of years
L E L no

>>717394305
A lot of the conversations about AI on /v/ feel like Skubposting. Reminder that /v/ is a skub board.

>>717391658
What the fuck happened to neural networks? Why does everyone want to use LLMs for everything?

>>>/trash/

Is anyone actually paying attention to this thing now?

>>717395860
A Neural network is just a more intricate configuration of LLMs. AGI is a meme, and governments better have a backup plan for when this massive market astroturf campaign fails and tanks the global economy.

>>717395994
>blaming LLMs for the shit crypto started

>>717396050
Apples and oranges. Both are memetech and we are well past the point where strict punishment of VCs and techbros is necessary for market correction.

>>717395860
The people that use it couldn't develop a neural network

LLMs are used by Low IQ people and browns

>>717395926
We just need the drawfags to find the threads again. But I am a little worried because last run we had drawfags already in at pallet town.

>>717396171
Stream has only been up for 10 hours
They'll be here soon

>>717396171
there's nothing worth drawing before he gets past the rocket hideout
unless he hallucinates something funny

>>717395926
I just pulled up the stream and there ain't shit happening. Looks like it's stuck sitting in a corner of viridian forest. I won't be surprised if it can eventually finish the game though. Pokemon RB is easy as fuck. Monkey with a typewriter and all that.

>>717391760
>Is AGI gonna happen in a couple years' time or not...?
according to every big brained coder yes

Did they ever fix the interface so Claude 4 can actually see cuttable trees?

>>717387447 (OP)
It's a very primitive and old model running locally on some guy's hardware.

He's just going to get stuck by bushes again, fuck retard claude I'm not wasting my time on him again.

>>717396690
opus 4.1 is a cloud model released just today

>>717391976
the term agi isn't even 10 years old

>>717396620
Probably not
But hopefully he did

>>717395860
LLMs are neural networks retard

>>717391760
LLM/Transformers/etc tech will never lead to AGI. There has to be a different way of doing things because it's not something you can brute force. It's like driving a car to the moon. This is something even Meta's big AI guy LeCun has been saying for years and he's paid money to pretend.
But virtually all funding is towards LLM and transformers so everyone is throwing money at something that can't be reached, and all the ignorant shareholders and venture capitalists will suffer severe burnout and buyer's remorse, pushing the tech back.

>>717396620
Was he able to see items before? He was very clearly able to identify an item on the ground an hour ago

>>717396967
lecun also said AI couldn't do convincing video

>The north exit definitely exists

>>717395994
The pain is going to be unreal.

>>717397072
Yes, he could always see items
But cutable trees just looked like normal bushes to him

>>717397687
lol no. it will be bad for poorfags but us investors and rich will make more money than ever.

>doesn't have to repeat all of the dialogue he reads
>doesn't respond like an overenthusiastic idiot when realizing what it is
Wow this version is light years ahead of the first
>still can't comprehend how viridian forest works
Nevermind

>>717387447 (OP)
it's retarded and a genetic dead end, like a chud.

>>717396920
It's been around for ages. Vidya example, X3 Terran Conflict in 2008
https://www.youtube.com/watch?v=nLKOzDvmoI8

HE'S DOING IT

Oh I thought that was the exit. Now he can get stuck trying to reach the void again.

>>717398206
consider me btfo

Gemini has beaten the game every single run btw

>>717393210
not yet

>>717396967
Lecunn thinks LLMs can't reach AGI because he has no internal monologue. He conveniently ignored the anthropic research that showed that LLMs actually understand concepts first and then language second, ergo at the deepest level theyre conceptual models. It was very important to Lecun that he never hear that because he's been pushing for his 'large concept models' for over a year now.

>>717398950
don't set off the sperg

>>717398950
good morning

Is he less retarded so far?

>>717393210
>>717399606
pls make lots and lots of zubrat findom content this time, thank you

>>717387447 (OP)
>a glorified auto complete bot can't do shit
shocker

>take text model
>make it play a visual game
>durr its not perfect
the fact it even somewhat works is crazy

>>717387447 (OP)
Gee two threads? >>717356413
>>717389584
>>717390112
(Posting from other thread)
Ackshually technickally claude 3 is claude 4, because the real claude 3 got stuck in an infinite loop within an hour into its run and had to be reset. We don't count this one anyway because it was too short.

>>717400298
>Ackshually technickally claude 3 is claude 4, because the real claude 3 got stuck in an infinite loop within an hour into its run and had to be reset.
I didn't even know that.

>>717400073
If he starts blacking out

I have to write a 15 page paper by tomorrow afternoon and I've barely started, what AI should I use for assistance?

>>717400760
gemini 2.5 pro on aistudio

>>717391760
I don't think silicon transistor based computers will ever do it. Maybe in a new computation medium, but I have no idea what that would even be.

>>717391760
easily
the sheer amount of papers and advancements this year alone proves it
also hrm might do it
https://github.com/sapientinc/HRM
tiny ass model that beats the fucking huge models already, once scaled it might be agi

>>717390112
bit of clarification here

the stream crashed in rock tunnel, but claude played on without an audience long enough to clear it and reach celadon

around 250,000 steps the game window display for the stream crashed in rocket hideout, but claude continued to play on behind it spinning and occasionally digging out for a break before going back in for more spinning

after a week of that bug the dev noticed and pulled the plug.

The only interesting thing to happen between the two crashes was a brief "perfume quest" arc where claude, unable to find the celadon cut bush, started to hallucinate the plot of the pokemon tv show and the always helpful CC bought into the delusion too

>>717387447 (OP)
Because pokemons games in Gameboy are designed for be complex mazes for sell guidebooks

Gemini won

>>717403606
Hopefully CC's gotten an upgrade too. It was pretty obvious during the last run that Claude was smarter than his own "guide," so any time the feedback chipped in while he was stuck it just made him even more confused.

>>717387447 (OP)
it doesn't have eyes, ears, a language, or wordly experience
it never developed a sense of reality and space and cultural norms and time like the children who play video games did in preparation of playing the game
it's no surprise that a computer program will have a hard time in this respect

>>717390924
AI bubble will burst and crater the US economy before 2030

>>717388757
Gemini actually had jeets putting inputs to guide it

>>717405773
Dont reply the the gemini bait anon... Its very clear gemini did nothing on its own + Its running 10 seperate agents.

BU

ZZ

>Oh wait that's the gym!
yeah this is claude alright

>>717395994
thing is I don't think AI has to really even deliver much more than it already has to not tank everything, look at some of the earlier big stocks there was a whole wave just based on SAAS products that are all absolutely awful and yet they are still around making billions

>>717387447 (OP)
I recently asked google's ai how many calories I had just eaten in a bowl of cereal and some scrambled eggs, and it gave the answer as over 5000 calories. It can't even manage simple maths.

>>717387447 (OP)

>>717395860
Neural network is the root technique.
LLMs are how you configure/train your neural network.

>>717387447 (OP)
I miss Sprou and Sand

>The yard awaits, Claude

>>717387447 (OP)
Because they are using text AI with image recognition instead of AI that was trained on the game

>>717387447 (OP)
>claude in big august 2025

>>717408998
We're coming to visit, ojii-san.

>>717409128
claude got an update today

>>717387447 (OP)
gemini beat it multiple times

>>717408991
Sand was one of the good ones

>>717409326
I wish I had a Sand of my own

>>717387447 (OP)
>Why is AI incapable of beating even the easiest of video games?
Because it hasn't been trained enough.
I'm so tired of explaining this shit to tech-illiterate digitards.
AI is not sentient, intelligent, or aware. It is a glorified calculator, but instead of using math, it's using abstract concepts, so it has to chart data for every possible interaction, when AI enters into a room in a 2D game, with a door at the bottom and a door at the top, it might stumble onto the door at the top, but generally it would need to chart movement through the entire room before it recorded which path was successful and which failed. It's not seeing the game, it's not understanding the sentence. It just takes in a formula (Prompt) and calculates the most likely result. But training it on videogames is different than training it on language because every sentence in a given language follows the same rules as any other language. Every game in a given platform or genre does not follow exactly the same rules so each game has to be trained from scratch with zero reference points.

>>717408042
Imagine being so stupid that either
a) you can't ask Ai a basic command
b) you lie about Ai being bad being you're... Insecure?

Which one faggot? I literally read your statement right to it and it gave me this. It just solved all of my masters' accounting questions on my open book test. Ai is going to destroy humanity. It isn't a gimmick. But it is destruction of education and knowledge entirely.

>>717407918
SAAS is an actual business model. LLMs are glorified search engines with none of the payout besides even less accurate information more frequently

How many steps will he take to reach mt moon this time?

>>717410946
4000

>>717410946
18000

How is little bro doing?

>>717390827
correct, what's your point

>>717411495
confused by doors

>>717411757
You love to see it.

>>717390807
Dev would have to redo all the tools for Claude to interface with the game.

>>717408991
BU ZZ has potential, I can feel it

>>717411996
Since claude won't make it to erika. Yeah. Bug type wont fall off before then

>>717409850
That fucker uploaded a blurry photo from his 100€ Xiaomi phone and expected Gemini to tell him "there are 100 grams of cereal which is 400 kcal". He's just a retard

yup... just a sliver of HP left... haha...

>>717391836
>I know no programming
and the core issue is right here. you are vibe coding.

>BU ZZ

>>717399892
LeCun has been right about some things but he's underestimating how far brute forcing will take us and how useful the models already are for making better models.
It's wise to keep a level head and to not get overhyped but he's gone past the threshold of healthy skepticism and he seems to now be in the denier bucket. I think it's because he's stubborn and wants to oppose Elon and the rest of them. But since Grok leapfrogged Meta, it's not a good look.

>>717412721
Vibe coding definitely improves productivity when performing high level tasks. AI is generally good at doing menial parsing style tasks for you

Post Zubrat femdom

>>717412715
red bar strats. the critical health noise cancels certain sound effects early, which speeds up fights. our brilliant icarus simply flew too close to the sun

doomer mode activated

>>717413552
I rather would lile to keep her as a daughterwife[\spoiler]

>I need to continue fighting with BUZZ even though it's hopeless
claude...

>>717413658
Daughterdom.

>>717409402
route 1 is right there bro

>>717387993
spbp

>>717391658
i really want to believe that this is a bubble and that the companies that keep trying to shove this garbage down our throats will go away
i do not need a fucking chatbot in a perfectly good website
tvs and washing machines and laptops do not need to have "AI" logos painted on them

BU ZZ'd out

>>717391830
I don't think the AI properly understands what is happening. If it's being fed screenshots every x seconds when the input is requested (which it seems to be, if the repeated pre-battle confusions are to be believed), it receives wrong info - spinning on tiles show it standing in a direction instead of what a child sees - smooth movement

B^U ZZ

>>717387447 (OP)
>he picked Charmander
Dig TM incoming.

>>717413690
With fat but firm tits in a latex bodysuit.

>>717413816
>i really want to believe that this is a bubble
looking at revenues vs capital expenditures for every AI company besides Nvidia, yeah, it probably is a bit of a bubble. AI capex contributed more to GDP growth this year than all consumer spending. Obviously not maintainable.
The technology's definitely not going anywhere, much like the dotcom crash didn't kill the Internet. But if/when there's a crash, it could be that we'll see less of companies trying to shove it absolutely everywhere, since a lot of that is performative shit for investors.

>>717414045
He will need to overlevel his starter to get past Misty. Charizard cannot learn Dig, so there's hope.

>>717414112
What do you mean? TM28 is possible to use on Charizard, HM02 cannot though.

>>717393210
Now, when you say "drained"...

>can't beat a game a five year old can beat
>it will replace programmers, you will see!! my gender studies degree finally will be better than yours!

>>717414352
it's back to manual labor for you, fag.

>>717414425
holy cope

Is anyone else finding the Claude bot runs to be incredibly blackpilling?

>>717414442
No, i have a good time.

>>717387447 (OP)
Because this particular AI constantly wipes out its own memory, instead of building up a coherent database.

>>717414352
take a look at Genie 3 then, and it's just the beginning
coding games and displaying graphics with traditional sprites and polygons will be relegated to novelty relatively soon

>>717414425
Well, Someone has to build houses for the government to give to the brown imports. AI won't do it.

Did the last Claude ever obtain the amazing TM02 before getting reset?

Anyone got links to any of the 'boxes from the old runs?

>>717413645

Left when Claude was jobbing to Brock's trainers, anything remotely interesting happen? Is he just grindan?

>>717414912
Yes. Going for Double Kick on Spike.

>>717399606
>>717400638
Finally /v/ is good for something again.

>>717414890
I'm so glad we got at least one typomon this run
it doesn't feel right without one

did he have a rationale for choosing charmander or did he just wind up getting that pokeball

>>717403179
@grok @t. knowers Is this true??

>>717415113
There is nothing rational about picking Charmander.

>>717393557
EoYprofits = profits *10
E=MC2*AI

>>717387447 (OP)
Token prediction is not abstract reasoning.

r8 clod III's team

>>717415517
>kino
Splash
DUX
Spike
>not kino
luna
wings
>not Sand
gust

>>717415517
wings MVP

>>717410946
Fuck you. I just lost.

>>717415410
It's entirely possible that our abstract reasoning is merely extrapolating information based on already existing information as well. Certain other quirks of our intelligence could be the result of biases that competition and evolution incentivizes.

>>717415305
yeah hrm got proved true

>>717393557
>if you look at microsoft's latest earning talk they give specific numbers on how much AI has saved their finances this last quarter
Post it? Does that come from anything but reduced headcount?
Their 2024 AI-related revenue was about $13bn of which $10bn was azure basically sold at-cost to OpenAI with preferential pricing, so not very profitable. They've spent around $135bn on AI infrastructure in the last 2 years.

NTA and I don't think AI is doomed or anything, it's clearly here to stay, but at present no company is actually turning a profit on it besides Nvidia. I think a bit of near-term skepticism is warranted on the financial side of things.

>>717415818
>>717415305
https://x.com/VictorTaelin/status/1950512015899840768

object permanence

It's over, the bubble will pop and it will make the 2008 crash look like a fucking joke in comparison.

>>717415947
>the bubble will pop any day
you've been saying that for like 4 years

>>717414539
I looked at genie 3. Unfortunately for Google I’ve also used YouTube vr. It’s so overfit that you could play their genie level next to some vr videos with 70% overlap. It’s super overfit and probably can’t do anything it doesn’t have a VR video to directly rip from.

>>717415820
I’m not a knower but I am old, and the LLM hype heavily rhymes with the exuberance when managers decided that communication technology was good enough to offshore all development. It will end similarly.

Schizokino is back on the menu boys

>>717415617

Puff board
Puff website

>>717415947
During dotcom tech companies had PEs of 500, we are not even close to the peak.

>>717391658
>The fact that these LLMs cannot learn or make complex connections
they do learn. but the ones you have locally don't as they use fixed models.

>>717414979
people constantly post ai porn of kids on /v/ though

>>717409663
I've been living under a rock for the past year or so have there been any cases of someone actually making a custom machine learning model (rather than just pluggin in, say, chatgtp) and training it on raw gameplay data for computer opponents?
The tech seems like it's getting good enough that you could get some decent AI opponents out of it (if not too good) but I guess that still depends on the game in question being played enough to have a proper amount of training data.

>>717416761
That's a bat.

>>717387447 (OP)
try to make it beat TLOU bro, pretty sure it'll be able to do it.

>>717416501
kino pic

>>717416984
you're brown

>>717391760
Couple of years no, either the current paradigm produces useful agents until 2030 or the AGI meme is dead until next century.

Why did they change the pokemon center message from "fighting fit" to "fully healed" in later generations?

brock pls why won't you talk to me ;_;

>>717391760
Unfortunately, the very debate of what constitutes "AGI" is incredibly autistic but vital to answering the question. If we go by Provost Zakarov's definition:

>"Begin with a function of arbitrary complexity. Feed it values, "sense data". Then, take your result, square it, and feed it back into your original function, adding a new set of sense data. Continue to feed your results back into the original function ad infinitum. What do you have? The fundamental principle of human consciousness."

Then yes, we absolutely can brute force LLMs into AGI. It seems like an exceptionally difficult thing to do though, however every time it seems like LLMs development is slowing down it has some ridiculous breakthrough.

>>717416787
AIs trained specifically on videogames end up pretty good at videogames, yeah.

>>717416501
Pure sexbait, irrelevant item-wasting party member

>>717391760
With enough power and chips , sure. But how practical is that? The economy doesn't work that way.
5-10 years for human level AI. It won't be LLM but LLMs will be used on training process

>>717391830

Turns out the human brain really is that hard to mimic and that we still dont really understand it

>>717387447 (OP)
Because it can't think. May as well just make it play back a demo. I remember the infamous blackout that pushed it to Cerulean City somehow, if I remember correctly.

>>717388249
Anon, it's Markov chains all the way down to quantum "randomness".
Also, civilization is getting destroyed by the people with such a massive degree of self-hatred, that they need to constantly spread it to everyone else to survive to the next day.

>>717419503
>Also, civilization is getting destroyed by the people with such a massive degree of self-hatred, that they need to constantly spread it to everyone else to survive to the next day.
/v/?

>>717392180
>sex
>zero human input
That's just masturbation with extra steps, anon

lass spotted

>>717419503
>Also, civilization is getting destroyed by the people with such a massive degree of self-hatred, that they need to constantly spread it to everyone else to survive to the next day.
And what's their end goal? Be the last one standing and then die miserably and alone?

>>717387447 (OP)
Because it's not Data's brain from Star Trek but in your computer.
It's a fucking text generator that doesn't even have a memory unless you continually feed it context so it can generate plausible sounding but often completely retarded output.
It's not self aware, it doesn't 'see' the world around in it's mind, it doesn't even have a mind.
It's a shitty program that glues words and expressions together according statistic probability.

>>717418446

>This is just NPC dialog about being Hot like Brock

What did Claude mean this?

brock fight NOW

retard is spamming leer instead of attacking lmao

SPIKE NOOOOOO

it begins

>>717422039
does he not have potions? also switching to buzz would save him, but he's too retarded

>onix is at 1 HP
CAM ON BU ZZ
YOU CAN DO THIS

BU ZZ VS 1HP ONIX

Come on BU ZZ you can do it

FUCK.

NOOOOOOOOOOOOOOO

SO
CLOSE

NEW CLAUDE?!!!
WHAT DID I MISS SO FAR?
QRD NOW
NOOOOW

Holy shit dude

no way

Claude4 is a loser.

I realized just before that I hadn't checked in on Claude's Rocket Hideout progress in a few days and booted the stream up and he's using a Charmander now so I'm guessing they killed the stream. I'm not gonna lie my enthusiasm for Claude just isn't here anymore but I'll always treasure the moments from the Mt Moon and SS Anne arcs

>BUZZ at level 3 is completely useless in this battle

>>717394670
no, you just need a different architecture
alphastar (deep learning) beat pro starcraft players years before llms went mainstream

HOLYSHIT HE PICKED CHARMANDER

Claude really hates BU ZZ

>BUZZ at lever 3 is completely useless in this battle
constipation detective

>>717422628
he pick charmongler
we're fucked

>>717422697
bu zz
it's pronouced
buzzey

>thought his pitiful damage was doing 20% of the total HP
>thought Geodude in yellow bar=1 HP
>thought Geodude still had HP when the bar was depleted
>thought Onix had used Tackle "8 times in a row"
Did he always allucinate stuff like this in battle mode (let's put aside the overworld)? He seemed saner in the past.

>>717422759
Fuck off CC i'm gonna call him BUZZ

>maybe I should have healed before the battle

>>717415305
it's a model trained on benchmarks lol
basically open book exam

tee hee it's tiem

>mt goon
>misty
Gonna be long and painful

>>717398950
Gemini isn't doing it on it's own.

>>717420417
based

charmander knows the way through the rocket hideout

Ah shit, here we go again...

>>717389991
Machine learning has been a subfield of AI research for decades. If your standard of AI is consciousness and at least as smart as a human then nothing is, but there have been journals and departments for AI long before this current bubble.

>>717424085
spike is my rock

>>717423596
How long before he boxes bu zz?

llms are bullshit tech unfit for any important tasks, but i think the one big reason they're never going away is that they have too many mass surveillance applications

they will fuck up all the time and a lot of people will fall through the cracks, but that's not going to deter any government lol

he's going to job to chadix again

>>717424256
I have started speaking nothing but lies to piss in the training data pool
even one flipped but can have catastrophic consequences on the long run

>>717424236
not this time

>>717424236
claude barely interacts with pc, especially not with an incomplete party

>>717422676
A big fat loser, and a total waste. Claude is a mess.

>>717424974
this is claude 4.1 you faggots

>>717424085
>>717424226
Did he already forget that he thought Spike would learn Double Kick at level 12?

>>717387447 (OP)
it's not AI

I think having prior general knowledge of Pokemon as a whole hinders the fun of this. If they could somehow extract the pre-existing Pokemon knowledge from the model and have it go truly blind without being able to cross-reference with information it pulled from wikis (which is inaccurate anyways because it keeps pulling information from future gens when this one doesn't work like that yet) it would be a lot more interesting.

>>717425308
The thing with doing a true blind play of pokemon 1 is that the game doesnt really explain what stats and the power of different abilities do.

>>717425308
the issue with that is that it adds a great deal more to the active context that can and will get forgotten over and over
instead of being a permanent part of the dataset
a small child isn't going to suddenly forget what type pidgey is after seeing it 500 times but claude will if it's going in there with absolutely no dataset to tell it that pidgey is normal/flying and what normal/flying is good or bad against

Brock 2nd attempt incoming

aaand we're back
will he succeed this time?

Reminder that Onix only has 10 (Ten) more points in its base Attack stat over Weedle.

>>717426616
This is the fourth attempt actually

>>717426623
Sure.
Hopefully he just clicks ember if it comes down to it.

>>717426679
The fuck

HERE WE GO

>>717387836
anon, a literal pet cat can get further than a journalist.
And the cat won't start meowing about there being too many dogs in the game

>>717392180
>sex
>ERP chatbot
ANON LMAO

>>717427084
If it's a proper journo cat, then it will be meowing about too many cats and not enough dogs.

>BLAZE 24/39 HP
off to a great start I see

Ok so the plan is... huh...
Pray he keeps raising defense, Ember ignores it anyway.

ENTER

>>717427717
EMBER*

Over status?

>14/39
BLAZEbros.....

>>717427758
we are jobbing again aren't we

>gen1 miss
Holy fucking over.

uhh bu zz sissies? how are we feeling?

>>717427947
he doesn't deserve this.....

WE ARE FUCKED

IT'S OVER
IT'S OVER
IT'S OVER
IT'S OVER
IT'S OVER
IT'S OVER

Not over...?

>He didn't change the pokemon to take Bide

>>717428181
the AI has no understanding of what bide does or how to play around it
anyway this is the clear

2 HP remaining on Blaze

Badge get

You doubted me?

>>717387447 (OP)
"AI" doesn't actually understand anything about what it's talking about or what is said to it. It just jumbles words together based on statistical probabilities, like a more advanced version of a Markov chain.

I thought AI was smarter and faster than humans. Why is this AI unable to beat pokemon after almost 6 months?

WON BIGLY

>AMAZING!
Alright settle down Claude.

>>717428278
I DOUBTED YOU

>>717428320
>ever listening to marketers

Claude got an upgrade? We back?
How much faster has it been so far?

Charmander not being able to learn dig significantly increases the chances of success this time. I'm feeling it.

Wtf is the LLM doing? To me it looks like its just the simple piece of code playing the game

After watching all this AI stuff.... I bought a few 10k of Nvidia stock and already more than doubled my profits on it. Should I sell now?

I think the previous best was 4763, so Claude IV is on the order of ~1000 steps ahead so far. Let's see how fast he blazes through Mt. Moon.

>>717428660
Brock was defeated 800 steps faster than Opus 4 despite picking Charmander as a starter

>>717428681
anon...

>>717428748
Considering their only plan to keep growing is to shill agents, yes you should sell. Agents suck dogshit and the market has over invested in AI when it's only real profitable use is for auto typing JavaScript code.

>>717428681
he's going to catch a diglett now because of your post

>>717429027
When should I sell?

>>717387447 (OP)
>BLAZE
>BU ZZ the beedrill
>SPIKE holding it down
THIS is THE RUN FOR SURE FAGS

>>717429191
idk whenever you want. Or you can hold. There will be some pullback but there will probably be another AI craze in 5-10 years when something new comes along.

>>717429447
>>BU ZZ the beedrill
[X] doubt
Claude is too retarded to actually train Weedle.

>>717422718
AlphaStar doesn't rely mainly on LLMs tho, right? And nobody would accuse it of general intellegence.
AI has always been useful, LLM is just not that.. Useful AI is a very complicated state machine. It has been used in robotics and trading long before the LLM trash has been popularized.
The thing is RL doesn't make normes shit their pants because it's not a chatbot. Reminder that we had facial recognition before LLMs and that all you need is basically openCV and enough CPU for everyone

>>717428320
>thought AI was smarter and faster than humans
It is in a way. But it's smarter in the sense that it's a conversational encyclopedia. It doesn't have a lot of shit humans have like spatial reasoning and self-awareness which help us actually solve problems.
>Why is this AI unable to beat pokemon after almost 6 months?
All that's happening here is Claude has been trained to output markdown that allows him to call functions called RAG. So it's still just a predictive text system but it predicts which tools it should call. That's why he can generate tons of text in perfect English describing what's going on, but he can't actually internalize and understand what to do many steps in the future.

>>717420417
who did you rotoscope for this image

>>717420417
thats a fuckig wall

>>717430273
it's the 'black guy talking to the wall' you commonly see in normie memes

>>717430805
>you commonly see
Maybe YOU do, but not me.

BLAZE took it like a champ, I'll give him that.

oh it's time to get stuck in mt moon for several hours again

Episode 4.1
Return of the Puff

>>717430805
I could tell there was a real human under there on account of seeing a few memes in my day

>>717432185
Will he pick up a goonslopper this time? Or is he going to get a Paras to fill in the need for a grass pokemon for Misty's gym?

Because this is the first time an AI has been set to the task of beating an early-gen Pokémon game.
With enough data, it will improve.
Then you will ask why it can't beat newer entries that don't use the old 2d system, and you will have people claiming that AI has hit its limits at last.

>>717432361
inb4 luna III

>>717431246
nice

It will never beat the best Sonic 1 human world record for another 100 years.

>>717432457
it's been doing this for a while now and still makes very rudimentary mistakes

>Badge already

It's pretty clear we'll have cleared Surge by tomorrow, Erika by Saturday, Safari Zone cleared by Tuesday, this is the run

>>717432779
Skill issue on the model's part, I would imagine.
If the top AI companies were tasked with convincingly beating a Gen 1 Pokemon game, it would be done in a week or less.

>>717431246
he had to carry this retard so hard

>>717387447 (OP)
Isn't it unable to see properly and mistakes a npc for walls? maybe the camera is set wrong. or is all that part of a llm too?

How long we gooning for this time boys?

>>717434224
only slightly longer than last time
i hope he catches a zubat instead of a clefairy this time

>>717434325
zubat is a common pokemon his autism will never let him catch it

>>717431246
Good stuff, drawfriend

>>717434371
so is weedle, and that didn't stop him

Mt Moon is near

>>717434471
he's allowed to catch weedle because it's like a classic pokemon thing you always catch a bug

is there somewhere else to watch this lil retard other than twitch?

Easy - Beats Lt. Surge
Medium - Cuts the tree leading to rock tunnel
Hard - Navigates through rock tunnel
Ultra Nightmare - Beats the game corner puzzle

Does he still have the hint system that the last Claude had?

>BU ZZ
kek
did i miss any kino?

>>717434591
Nope, but you can use Streamlink if you don't want to deal with Twitch's website.

Fuck that made me remember I started drawing a Zubrat months ago and forgot about it

used Potion on Spike

>>717434961
Not sure. I haven't seen Claude mention Oak's parcel randomly. If he starts randomly mentioning ledges on Route 4 or the Trashed house in Cerulean then he does still have the hints.

pick your lass

How can AI win a math competition but it can't beat pokemon? Preposterous.

>>717391079
only a humanities major could believe work efficiency is something you measure with vibes

>>717435735
always bet on the burgandy haired lass
cute lasses anon

>>717435735
4 > 1 >>> 3 > 2
Blue hair is the goat but brown hair is too cute

>>717434325
>spoiler
we’ve been over this a billion times. Claude gets hard for RARE pokemon, not shitbats
You LOST the first, second, and third time. Only time he ever caught one was to fill out his pokedex and immediately boxed it. Zushat is just not a good pokemon in gen 1 stat wise and movepool wise. That line REALLY needs crobat to even it out.

awww shit, charmeleon incoming

>>717391487
He has actually cleared that maze a few times. Even entered the lift once. It's B3F that he can't solve because he gets lured in by stairs he can't reach and then gets trapped in his goon spot until he gives up and uses Dig to escape.

>>717431246
Oh my god lmao

BU ZZbros, Claude wants to box us...

>>717436067
You're scared. You can smell it.

>>717436960
Updated

>>717437185
BU ZZ probably won't be with us much longer because both Claude and CC concluded he's completely useless.

>>717435735
OUGHYE, BROWN HAIR
4 > 1 > 3 > 2 (who the fuck)

>>717437297
I'm aware, don't worry

Just when I thought I was out, THEY PULL ME BACK IN AGAIN
Welcome back spike lol

>>717393326
average 4channer reading comprehension is on par with that typing ability

Kino is BACK

>>717413816
>tvs and washing machines and laptops do not need to have "AI" logos painted on them
Yes they do, I want my washing machine to have an anime avatar appear on screen and talk to me.
>I see you need your clothes washed today Goshujin-sama. Just drop them in my mouth Itadakimasu, NOMNOMNOMNOM! Arigatoooooo.

The stream got unbearable to watch towards the end. Has he been updated with anything or is it a reset to get more than 10 people watching the stream?

>>717438418
Model upgrade to 4.1
Full reset

>>717437958
AI has nothing to do with having an anime avatar you porn addicted tranimefaggot

>>717438418
Opus 4.1 launched but all it's improvements focus on software development and none of that shit is going to help Claude.

>>717438749
Thanks

>>717438805
it's all interchangeable, transformers are fucking bullshit. it's not going to be as impactful as if they made a model for playing video games, but it'll still improve it

>>717435735
Those Roman Centurions certainly look cute.

>>717437958
how about a central hub in your house taht has the AI and then the other appliances just have a port from which she gains control?
granted if you want an AI for each appliance then that is another thing.

If this is Claude 4.1, what numbered Claude do you think will eventually reach the end?

>>717439139
I have a good feeling about 7.83

>it looks like a pidgey
You DUMB fuck claude, that's clearly a spearow, you've seen 8000 of them the past year you retard faggot

>>717387447 (OP)
Idk but didn't a literal fish play through this game a few years ago?

He still has the hints until Cerulean

>>717387447 (OP)
it would be very easy to train a neural net to beat pokemon, Im pretty sure it has been done already
using an LLM for this is like asking why your washing machine is so bad at cooking your food

>>717391760
Quantum computing has made almost no progress in like decades last I checked

>>717431246
Keep carrying the thread anon

>>717441027
I'll see what I can do, hopefully this run delivers funny stuff to work with, and won't just be a carbon copy of the last run. I don't wanna repeat myself too much.

>>717420270
To feel better about themselves. Also, to get more political power.

The biggest "justice"-proponents of ANY kind are usually the worst perpetrators in that context. No matter the context.
For example "climate justice" is when the beaches in the resorts are finally freed of low-income peasants, that ruin vacations for the well-educated climate angels trying to save the planet.

>>717432882
Or OR it will not complete the game and die off every time it enters the champion road. The AI is too stupid to understand healing and strategy.

>>717420357
>It's not really thinking! It's just reacting to what is happening around it based on short term memory/context combined with long term trained data merging similar experiences and patterns to guess the best thing to say and do at a given moment!

>he's not catching jigglypuff

>>717440730
Its been slow going. Google announced provable quantum supremacy in 2019, and some pretty insane chips, but its been quiet since really.
Theres a lot of work being done on the application of quantum computers rather than the continued development of them. For example, they can already easily break a lot of encryption algorithms, but by 2030 they think it'll be able to break any form of encryption within minutes.

The actual bleeding edge of quantum computing is supposedly all being done by China, especially now with their huge energy wins in solar, wind, and now their fusion reactor. The "artificial sun" which in testing was able to sustain stable plasma generation to redirect as an energy source for over 1,000 seconds or some such.

>>717441769
At least no Sand either yet

>>717387447 (OP)
LLMs aren’t AI and you’re all fucking stupid

>>717442150
none of it is artificial intelligence it's just the easy blanket term used to describe any of this

>>717441702
>dude humans are le same xD
Reasoning and logic is different from short term memory you fucking retard.

>>717436960
>>717437185
blaze is now among the all time top 5 coolest pokemon

>>717442150
>>717442228
I sincerely hate whoever popularized the term "AI" for this shit. Its not even easily filterable because "AI" is in a fuck ton of english words

>>717433131
>If the top AI companies were tasked with convincingly beating a Gen 1 Pokemon game, it would be done in a week or less.
how convenient for your belief system

>This thing that confidently lies and makes shit up and cant parse sarcasm will totally be AGI, 2 more weeks saar
AIshit would be much more tolerated if the biggest proponents of it didnt constantly try to overhype the fucking shit out of it and then get buttmad when people call out bullshit.

>>717442843
The hype must continue or the bubble will burst

It's time for Mt. Moon

>>717437185
the claudes slowly but surely forget their spikes.

>>717387447 (OP)
Because it's in the name ai. Artificial intelligence. It is not actual intelligence and it never will be. It's a tinker toy piece of shit at best.

claude...

>>717443443
Wow he beat Mt. Moon already!

>>717387993
>>717394670
no.
the problem is two fold
>1) Claude doesnt see a live feed of the game,
he takes screenshots and interprets them against a map of what tiles can be walked on.
This causes a number of issues that are difficult to work around, on top of his poor image interpretation at low resolution
>2)Hallucinations
When Claude begins to get "frustrated" by a lack of progress it tends to get schizophrenic as it attempts to think outside of the box
>2) Context limitations
The fix to the above problem means Claude struggles to build a long term memory of certain kinds of puzzles.

He has, on his own, repeatedly figured out HOW to solve the more complex puzzles of the game like the arrow mazes, as well as figuring out their behavior despite being BLIND to the entire process of the arrows moving him around.
But he never gets to keep it in memory, because of the safeguards to keep him from going crazy.

Route 4 North... Home

>>717443817
ah fucked up my numbering.
oh well, CC can wipe it from my memory

did we finally make it out of mt moon?

>The feedback suggested I already went through Mt. Moon
CC YOU DUMBASS

>>717444392
Already gaslighting
We are so back

>>717442843
Its all a big tech corpo bubble thats on the verge of bursting constantly so they have to invent new made up shit constantly and hype the fuck out of it.

Nearly every major F500 company has massively bought the hype, collectively, trillions have been spent on research, integration, etc. Now that its proving to be more a detriment to engineers and what not and people are starting to become aware of how ineffective it actually is, the propaganda machine is in full effect. To me, it just seems like an enormous wealth transfer scheme.

>>717390529
we had full logs that started before the OBS crash as well, AND claude was getting mere inches from the exit on stream before the crash.
He basically got lucky once and his paranoia failed to make him run away.
this same thing would happen repeatedly on stream while he was stuck in rocket corner, where he would fuck around long enough without wiping his context or running to start exploring the right way and halfway solve the b3 maze
it happened like 4 times over the course of 3 weeks

>>717444721
We have management walking around the building begging people to use AI more because none of the engineers like it at all kek.

>>717439070
>how about a central hub in your house taht has the AI and then the other appliances just have a port from which she gains control?
>granted if you want an AI for each appliance then that is another thing.
Yeah I could have something like that for the entire house, but it would be like the head maid and I would want it to be really strict and no-nonsense and verbally order the others around to make sure they aren't slacking.
I don't want a central point of failure and I would still need each appliance to have slightly different personalities. Like I would definitely need my microwave to be a tsundere.
>I'm not cooking this because I want to baka goshujin-sama!

>>717444875
Nobody on my team uses it, pretty much ever even though its mandated. Even the stuff you can use built into your IDE if you have one is near fucking useless 90% of the time, even for basic scripts. What happens is you spend more time managing and correcting AI doing the work for you than doing actual work, but that makes you hilariously inefficient because getting it to do something right and write a line of code that isn't completely broken or inconsiderate of the rest of the program is near impossible a lot of the time.

And yet we've got every middle manager, director and CEO forcing it on us 24/7. Remember, AI = actually indians, thats where the jobs are going, not to artificial intelligence, indians.

>>717443443
>>717443781
>>717444392
Intelligencia

>>717433131
>Skill issue on the model's part
I want you to know that you are fucking retarded and you have no idea how this shit works.

>>717444875
My IT company made their own AI chat to sweep through our ticketing system to answer questions.
Every meeting its the same
>Guys, you need to use the AI chat WAY more

Do you think Claude will try to murder people again this run?
Watching him try to cut the fat guy down in Celadon was funny

>>717444392

>>717444721
>To me, it just seems like an enormous wealth transfer scheme.
The next ultra rich influential guy will be the one whale investor that cashes out of AI at the optimal time and creates the panic.

He's in, let the gooning commence.

HE'S IN

>2 steps in
>zubrat encounter
Good start

this will be your home for the next month and half, get comfy.

>Seen any Zubrats?

thank go for claude threads. Rest of /v/ is just anger or porn/porn-adjacent posts. These thread have also a lot of lewd posts but only like half, so it's fine

>>717446948
new oc? good stuff

please, just a drop of exp for BU ZZ...

>>717447716
Nah, I made that months ago. I have a feeling I'm gonna be reposting a lot since this run's probably gonna be mostly the same lol

>>717442150
>it writes like human
>it talks like human
>it does math like human
>it codes like human
>it draws like human
>it passes turing test

>>717446917
Nah out by Sunday easy

The ladder/stair confusion remains, the model is as blind as it has always been.

>>717387447 (OP)
Because the goals of games are incredibly abstract

>>717448931
nah if you could see his current ladder connection matrix ascii diagram you'd fucking shit yourself

>>717448931
LITERALLY the exact same sprite, wtf Gamefreak?

>BU ZZ (to be boxed)
bugbros, I'm not feeling so great...

>>717449219
it's for the best, I don't wanna see that lil nigga getting squashed because of clod's retardation anymore

>>717449219
BU ZZ will get to sleep peacefully while the rest end up in hideout hell once more

>>717449127
put it in grayscale and blur the fuck out of it to account for ocular resolution

>>717449830
is this how clod truly sees....?

>>717431246
The king returns

>oh fuck, we're so ba-
>Mt. Moon

GOONSLOP DETECTED

RED ALERT

>>717451280
HAPPENING CANCELLED

>runs from clefairy
goddammit claude, I need my cute pink slot waster NOW