Google CEO praises their AI for beating Pokemon

Google CEO praises their AI for beating Pokemon

How legit was this run?

what happened to claude plays pokemon

Big deal, I bet it when I was 7 and I could barely locate my dick back then

It's just advertisement

Can't Nintendo sue Microsoft? I doubt they had their AI run a legit copy of the game on official hardware.

Didn't they give hints to Gemini when it was stuck?

Gemini was the one where the dev constantly helped though

AI always needs training, so I'm not surprised.
Development takes time, AI development has taken a different shape somewhat.

The game isn't done until you catch Mewtwo.

barely, he was feeding it """hints""" the whole time

Haven't played a Pokemon game in like a million years, aren't they easy as shit anyway why is this an achievement?

claude couldn't cut the tree

These models don't learn though. Here it was the equivalent of the dev putting his toy car back on track and saying that it didn't count that it went off course. It made the whole challenge pointless.

Check out claude

Still in Mt Moon with the fat bird being 30 levels higher than everything else

Business as usual

whatever happened to claude's gold adventure?

Livepatched instantly whenever it struggles remotely with anything

it took 7 year to develop skill for you, you will die and lose skill after 80 year
AI do not die and only improve, think of all skill ai can get in 200 year in time normal man die and keep

Because their competition, Claude, is actually mentally retarded.
It's just Google boasting that their AI is better than Anthropic's AI.

gemini had a fully visible map at all times and not just its visible range so it didnt work off its own memory of the layout but rather meta knowledge, which invalidates the run imo

What does Microsoft have to do.witht this?

Claude had meta-knowledge too. There's pre-existing information about Pokemon in its data set.
That's why it would do stuff like get type matchups wrong, because it was using information from later generations.

Both AIs "cheated", yet Gemini beat the game and Claude didn't.

could claude beat nu-mon where the player gets railroaded and side characters explain everything they should do?

having map knowledge all the time that you havent even discovered yet makes it easier to navigate to your goal without getting tripped out

Now do it without training wheels or dev input.

Emulators have been ruled legal.
Format shifting has been ruled legal (that'd cover that'd be ripping your game cart to a ROM file.
AI scraping content has been ruled legal.
What do you valid complaint do you expect Nintendo to have?

CLAUDEBROS???? HOW DO WE COPE???

gemini doesn't have a pidgeot with gigantic tits so i don't care

What do you valid complaint do you expect Nintendo to have?

Fuck you we're Nintendo.

Claude has tiny context and is kinda retarded, it was only good for coom back when Sonnet and Opus weren't molested by Antrhopic to turn them into bad coders and targeting software to bomb Palestinian children.
Gemini in comparison has very, very large context sizes.

AI that requires the electricity cost of a town, to beat a game that required 4 AAs to operate

I asked Gemini why his win had any importance, his answer if anyone cares

Think of future potential from training ai to solve harder and harder puzzles in future ai will cure cancer with no one dying from how good it become at solve puzzle

Just got Android 15 the other day. Anything useful Gemini can do? There's a girl at the office that uses ChatGPT constantly but I never do anything with AI beyond asking questions that are a little too complex for a google search and a little bit of image gen

Good morning sir :)

And all it will need is a quantum supercomputer powered by a couple suns?

Better than an actual drooling fucking retard singlehandedly tanking the reputation of a decent LLM because he's too unintelligent to actually craft the fucking instructions in a productive manner.

Why did Claude have thousands of people watching it bumblefuck around forever making no progress yet Gemini beat the entire game with like a dozen people watching?

moon landing was real, guys! source? trust us ;)

A 10 year old can beat Pokemon and you don't need to craft any fucking instructions

Gemini was literally helped out and handheld every step of the way by the overbearing dude who made it. It was basically him who beat the game on autopilot rather than the AI actually taking the steps necessary to do it themselves.
Probably in the same megaloop he was in when I last checked two weeks ago if it wasn't mercy killed (again)

It’s still in Mt Moon.

Still? Christ they should just pull the plug already.

Yep. He's made it through multiple times in the past, but that was before DIGLETT.

not only was gemini rigged by the dev, but it also had no personality, claude gave us peak schizokino like ONLY SAILORS KNOW HOW TO LEAVE.

All these tests are meaningless until the input is only a matrix of pixels with no RAM access bullshit, and the knowledge base ensures no previous knowledge about Pokemon is present.
At most give it the game manual along with its general-purpose knowledge, but nothing related to Pokemon.

I think it's good at making porn and little else. I'm gonna need a little more convincing to believe it's worth the untold billions being poured into it

easily fooled by smoke and mirrors
lazy and retarded and prefer to take the easy way out always

AI will make India a superpower for real.

Oh wait, it's good at translation as well. That's definitely something.

It's almost all entirely a grift. It replaced cryptocuckery/NFTs as the go-to buzzword for con-men and smoke vendors. The zoomzoom who made Claude play Pokemon Red is at least using it for something mildly productive, a thesis of sorts that can show off the potential of LLM problem solving.
Just wait for Quantum Computing to "take off" it'll replace AI in the modern discourse fucking INSTANTLY

I was hoping he would level up DIGLETT enough to learn a new move and somehow manage to overwrite dig.

It's not about the model. All current models are more or less similar to each other. It's about the prompts that they feed to the models to play the game. The Claude prompting system is trash down to the color scheme used for the tiles, and the Gemini one was less trash and the devs were feeding it hints.

Gemini in comparison has very, very large context sizes.

Gemini had a human curator who was constantly telling it what to do with prompts. Whenever the faggot would go to sleep, gemini would literally stop doing anything at all. It's run was literally no different than playing the game yourself but in a more roundabout cumbersome way. I wouldn't even be surprised if the guy running it was a jeet.

overbearing dude

no fucking kidding. the guy was constantly posting updates on the claude reddit page and bragging every single step of the way. he was treating this whole fucking thing as some kind of race and probably feels like someone owes him a trophy. I genuinely don't think he even understands what the point of this experiment was.

I read a story with that premise and it ended with the AI trying to kill itself.

They can't create

Because everyone who was aware of these two playthroughs understood exactly what was happening. I'm sure the google CEO or whatever has absolutely zero idea what any of this actually means.

How retarded

He was probably also the one personally posting Gemini updates in the Claude threads. Anon Babble has a lot of contrarians and autists willing to dedicate themselves to one hyper-niche style of annoying ritualposting but I doubt a single person in the Claude threads was really here to stir up shit between the two algorithms. Gemini's run was astroturfing from beginning to end.

The problem is that it's using a logical conclusion to solve a problem that makes sense on the surface, but will never allow it to progress. It knows that when you get stuck in a cave, you can use dig to get out. Makes sense. It doesn't realize that using dig takes you to the last pokemon center you visited, which in this case is the one right before Mt Moon. It will always get stuck, always use dig, and always reset any progress. There's a chance he could be much further along, but the very instant he thew a pokeball at DIGLETT, the entire run was compromised. There's a similar pokemon center right outside of Rock Tunnel, and if he ever made it there with DIGLETT in tow, the exact same thing would continue to be happening there as well.

Have you ever jingled keys in front of a baby? Same principle.

I feel like they're just two dry frogs in a situation.

It'll all be Jeets and Chinese.

Maybe they're making jeets CEOs because CEOs don't actually do anything important and you can pay jeets less.

Ever seen Memento?

Thank you mr creepy Russian man

Gemini in comparison has very, very large context sizes.

The "context size":

I worked with an Indian guy. Yknow total business style NPC's who spend their lives on LinkedIn? Yeah it's that. No idea about how or why things work but with an insatiable desire to demonstrate that they do.

i hate that this thing and claude both used the gen 2 graphics patches, romhacks are so cringe.

A new advancement in technology happened.

Tendies: LAWSUITLAWSUITLAWSUITLAWSUITLAWSUITLAWSUITLAWSUITLAWSUITLAWSUITLAWSUIT

Why this is the first thing they think about?

So this is the benchmark for AI, huh? Beating an old Pokemon game? Why not more modern games?

OG pokemon looks like shit and I grew up with it.

this was done for a practical reason, so that the AI could see that things were different colors. This may or may not have been useful for Claude. Gemini went a step further and changed the pokemon sprites as well because they wanted theirs to "look nicer". Using the graphics and color patch was useless for them because it didn't need help seeing anything. It was entirely curated by a human who was watching it and telling it what to do, so in their case it was only used to make it look nice and not for any other reason.

I don't think the streamer was even pretending it was "legit".
You do a full run cheating when necessary. Then, you look at the parts where you needed to cheat, and see if you can modify the starting conditions to account for those situations without cheating.

The models are likely just not powerful enough to win at this point anyway, until they're able to manage self-made memory files effectively enough that the 'psuedo-learning' idea actually works.

I don't think the streamer was even pretending it was "legit".

He was literally bragging and mocking Claude every step of the way and treating it like a competition between the two. Nobody who was following Claude even necessarily cared about the stream. It was all about the in jokes and the OC it spawned. The retard doing Gemini either didn't realize that, or he was hoping to spawn his own community around Gemini. He doesn't even work for Google or Gemini, so it was literally either just some autist or jeet who wanted the anonymous glory of being the first to curate an AI to beat pokemon. Notice how he picked Blue version instead of Red like Claude. It was never about proving anything, it was always about trying to find clout where there was none to be earned.

Some anon posted a thing about how jeets are obsessed with being over others and giving orders to them and AI fills scratches that itch by letting them feel they're ordering around a subordinate to do all their work.

What agentic AI is trying to accomplish is to have the same reasoning and cognitive skills of an actual human.
We develop things like object permanence, the understanding that just because we cannot currently perceive something or are not receiving any stimulus from it does not mean that it has ceased to be.
There's a vast number of such concepts that an AI system (which has multiple models working together), that need to be developed into the system for it to achieve the hurdles that are being identified in these trial runs.

Claude was treating blackouts as a reward because it was taking them out of mount moon to a safe space and fully replenishing their team's health, but was unaware of the money penalty that incurs and does not learn that it is being punished accordingly.
An effective model needs to be able to assess that the money that it had is suddenly reduced or not consistent with its earnings from trainer battles or previous shop visits, to identify that something's amiss, and then reason why that money is missing.
The information that the original Pokémon Red and Blue conveys is at times limited and relies heavily on context and awareness of your previous actions, such as how a back door in some houses is outright not visible on the world map and relies on you remembering the passageway. The older games had communication shortcomings like not telling you that you've lost money, which newer games make sure to let you know that you dropped some money when you hurry off to the nearest Pokémon Center.

jeets are obsessed with being over others and giving orders to them

I dunno man that seems like a pretty universal human desire to me

There were a couple instances where claude would check his money and be confused about why he had less than he expected

Proofs?

Very good saar! You earned your $10 Google Play Store gift card

hey what's up geminigger, did you enjoy the Google jeet giving you a shout out?

Isn't quantum computing a meme that's two decades old yet is always touted as the future? Same with carbon nanotubes

It's touted as the future because it's theoretical at this point

Look at how many different technological breakthroughs have been announced in the last 20 years and yet have literally never left the lab. Either they're too expensive to be practical for commercial use, or they were just lying. In the case of China, it's always a lie. They recently announced they achieved nuclear fusion recently or some gay shit, but they've been saying the same thing over and over for decades.

jewgle

naw

how the fuck do they even justify this. if they're going to make their AI do one but not the other, they REALLY need to be transparent about why they're doing it. the only thing this tells me is that eventually people are going to use AI to answer political questions, and faggy leftoids will use it as proof that leftoidism is the correct answer to everything while denying the idea that the AI is very heavily curated by a bunch of faggy Sweet Baby types ran by fat black women.

What did Gemini do right that Claude did wrong?

Gemini had a jeet telling it what to do every step of the way. Without that, it would probably be stuck in Viridian Forest at the beginning.

how the fuck do they even justify this.

they dont need to justify it because they know they will get away with it

Claudefags in shambles

Claude was dropped into Pokemon and told "Go beat the game" without any extra help, and did surprisingly well outside of his terminal blindness. This was fascinating to watch as a way to understand the strengths, weaknesses, and current state of LLMs. Gemini was the dev playing Pokemon with AI to handle the minutia like actually wining battles with the level of handhelding he got. No one really cares, nor is impressed, by Gemini being able to pick the correct moves in battle.

without transparency on the development side, AI and chat bots can never realistically be used for anything intellectually honest, which I guess is fine, because leftroons are the most intellectually dishonest people on the internet. Someone told them that conservatives are stupid, so now they all think they're smart because they do exactly what liberal media tells them to do without even thinking about it critically for one second.

explicitly trained to hate white people and worship niggers

im curious to know how much energy it took to finish the game compared to a person

Theoretical? There are actual quantum computers running quantum algorithms right now. They are not showcased because they are really good at solving some very niche problems, with barely decent improvement on time-complexity over regular computers; but completely shite at anything else.