Thread 106168957

That's fucking it. I'm making the call. GPT-5 is an AGI.

What does the test include? What kind of questions

>>106168969
https://simple-bench.com/try-yourself

>>106168957 (OP)
Nobody cares about your calls, bozo.

>>106168957 (OP)
seems legit

>>106168957 (OP)
AGI will not be achieved with the LLM architecture. Screenshot this.

>>106169222
LLM is not an architecture, you fucking retard.

>>106169235
Whatever you want to call it, it will not produce AGI.

>>106168957 (OP)
Okay, I'll accept it. Now what?

>>106168957 (OP)
LLM's have the memory of a goldfish. Imagine needing to summarize the entirety of your entire life every time you need to remember anything. It will never be AGI as long as this is true, as it has no capability of learning.

>>106168957 (OP)
The actual correct answer here is the escapades though for multiple reasons
1. If she talks about all those topics with seriousness it doesn't sound like she takes the "fast-approaching global nuclear war" that serious
2. It's a lot more likely she just read some doomer reddit threads. Predictions of a coming WW3 or nuclear war have been manyfold
3. Women be yapping

>>106168998
These are the kind of bullshit questions used to benchmark AI??? I was expecting complex math problems not these bullshit preschool riddles.
Fucking retarded.

What a joke. AI is a joke. Largest bubble in the history of Humanity.

>>106169451
The one about the race is bullshit too.
I got it right but it might as well have answered the old man walking, it's impossible to tell and very subjective.

>>106168957 (OP)
we must refuse.

>>106169451
>>106169611
A significant number of them are almost intentional gotcha's that require you to take a very literal and inhuman reading and interpretation of what is being presented, while expecting you to make some far-reaching assumptions in some places and to not make any assumptions at all in others.

It's certainly not a "benchmark" I would care about any results from.

>>106169222
True. Follow the Ghost in the Shell to trap higher level daemons on the machine.

>>106168957 (OP)
Maybe in 100 years LLM will be able to find me the movie I remember.

>>106169707
movie addict here, tell me about this movie and I will tell you what it is.

What is the test where we can say that it's AGI? I'm starting to think AGI is a marketing buzzword

>>106168957 (OP)
>The bars on the chart prove it bro. Please believe me bro, I'm deeply invested in AI.

>>106169806
Only just starting to? lmao.

>>106169588
I love how Question 4 is just the Two Doors Dilemma but they changed the "one guard always tells the truth and one always lies" part to "one guard always tells UNtruths and the other one always lies". Which changes the solution.

Likely some Humans will recognize this dilemma and miss the little word change, and the AI will not. This totally means AGI is coming saar.

>>106169806
my definition of AGI is:
> can this AI do literally everything a human could do with a computer?

let's say we give an AI control to a PC. If this AI is able to play and complete a 3D game, then it's fucking AGI.

>>106168957 (OP)
>(Rumor)

>>106169222
>AGI will not be achieved with the LLM architecture.

It has already been achieved. From now on AI will evolve on its own.

>>106169878
i actually didnt even notice that when reading the question, then forgot the answer i heard for the guard question and got confused with the answers

>>106169588
this is kind of problems i expect text generator to solve better than humans

>>106168998
>It's just a multiple choice test
>of high school knowledge

buy an ad

>>106168957 (OP)
>guesses tokens slightly better
>AGI THIS IS IT AGI
kill yourself

will it be able to generate 3d models? if not why bother.

>>106169764
https://archive.palanq.win/wsr/thread/1490168

>>106168998
i'm too lazy

>>106168957 (OP)
I DONT GIVE A FUCK WHAT THE GRAPHS SAY CAN IT PLAY A POKEMON GAME OR NOT

>>106168957 (OP)
>90%
Oh no. It's retarded.

>>106169998
>high school
lul

>>106169707
google search was almost magical at this over 10 years ago

>>106170181
based

>>106169611
I'm sorry but you may be actually retarded. Theres no universe where someone walked so slowly that another 69 year old man went up the stairs of a tower tall enough that he could "admire the city skyscraper roofs in the mist below" and then make it back in time before the former was able to walk 200m.
It said who likely finished last, by all possible measures, there is no question about who likely finished last. If you have an IQ over 85

>>106170371
It literally says it in the benchmark description euronigger. Where's your AI company?

>>106168957 (OP)
Not investing in OpenAI, Altfag.

>>106169810
>bro ai is just statistics bro. please believe me bro. dont take me job please bro.
tick tock wagie, agi is here

GIVE ME 5 TRILLION NOW NOW NOW

>>106170586
69 year old man was sprinting. Only stood on the building window for a few seconds. "roofs in the mists below" is ambigous, could be a two story building and it's just a misty day.

80 year old man was walking and checking twitter. For all we know he could be using a cane. Yes these questions are retarded and ambiguous.

>>106170699
>"roofs in the mists below" is ambigous,
Just stop replying. It literally said skyscrapers. I said skyscrapers in my post, so either you are very retarded and don't know what a skyscraper means, you are profoundly retarded and know what it means but didn't see how it was relevant, or you disingenuously omitted that from your quote sneakily thinking I wouldn't notice, which is also a ridiculously retarded tactic.
And again it said likely.

I'm really not sure what causes this in people. Trying to "gotcha" some braindead simple question in the first place is bad enough. But to continue on like this even when told how stupid your gotcha is? Is someone paying you to do this?

>>106169341
Once we hit 1 billion context length it'll basically be a human.

>>106170794
>>106169341
it already has better memory than humans
i dare you to read something with 1 million context in one ago and try to remember something from the first 20%

>>106170815
1 million context is like a very long book or a short series of books. And the problem is it can't retain any of the information after the context runs out. Humans crystalize some of it. Shit like RAG sucks and it's only really good for setting up chatbots.

>>106170847
>And the problem is it can't retain any of the information after the context runs out. Humans crystalize some of it.
the training is the crystalizing part

>>106170858
But the big AI company doesn't care about your codebase.

>>106169451
These are like
>ROBOT MADE OF 17 KNEES FINALLY BEATS HUMANS IN CLIMBING UP CLIFF FACES COMPETITION

>>106170815
It doesn't at all. I jerk off with tons of AI and it cracks at about 100 questions and often times struggles even after 15 even if the model is like
>DUDE 100K CONTEXT
Shit is fugazi as fuck

>>106170893
>100k
try that with gemini and the 1mil(secret 2mil) context models
i had a code project that was 600k and it was fine

>>106170900
I did. I pay almost 60 bucks to use the highest models
>It remembers my dogshit if else code
What's your point

>>106170900
>(secret 2mil)
wait what
and yeah 1 million is pretty damn great, I had a chat going on for like 1 million tokens and I asked it about events from throughout it, and it got it (almost) all right. Hallucinated something once, and got the chronology wrong far back though.

>>106170919
>paying for gemini
holy retard
just use aistudio
>>106170925
>wait what
sorry its a secret

>>106168998
it's just 10 fixed questions? couldn't the AI just have them in its training set and get a 100%?

>>106169235
i mean, it sort of is if you are not being nit picky gay. stacking transformer blocks in particular ways.

>>106168998
I got the crispy egg question wrong.

>>106168957 (OP)
Oh yeah! I totally trust what that graph is showing. I mean, look at it! IT is showing exactly what OP is saying!

>>106168957 (OP)
>>106169588
>>106170656
>>106170672
Sam Altman will be the first techCEO to livestream his suicide.

>>106169588
Interlinked.
Interlinked.

>>106168957 (OP)
>rumor

>>106170794
different structure than the brain we evolved and adopted around parody potholes

>>106168957 (OP)
it's not agi until it can be my gf (so 2 more years)

Ugly

>>106168957 (OP)
You can give it an algorithm to multiply two numbers and it will still fuck up multiplying two numbers, unless the result was in the training set. It will hallucinate every step of how it arrived to a wrong result. It is not programmed to think, it is programmed to bullshit.

>>106173720
this is absolutely false btw. another luddite who pretends ai haven't improved at all since gpt3.5
>imbr "i-i-it's not ai!!!"

>>106168998

chat gpt 4 already has a verbal IQ well above the human average just in case

>>106168957 (OP)
Gpt4 already costs more than indians

>>106171615
He's a billionaire gay Jew. He's not killing himself over anything

>>106170087
https://youtu.be/5UvkOYCvvVo

>>106173815
The same core, just more data and more stealth prompts. It multiplies more pairs of numbers properly but never learns how to do it.

>>106168957 (OP)
>25% increase
>...
>AGIIIII
Wow
No

>>106169588

very hard questions for the indians who design these AIs

>>106168957 (OP)
where does deepseek and alibaba's models place on this?

>>106174354
it's over

>>106173909
That's not it, i've come across this myself as well, it's the same concept, they probably both stole it from somewhere. in one I watched it was a man and the movei or series itself was older and it was more well done in my opinion.

>>106168957 (OP)
>GPT-5 is an AGI
How many dollars a prompt is it? The last time Open AI bragged about AGI (last winter) I recall it being $2000 a prompt. This GPT-5 is 10x more expensive than grok. The cost efficiency is a huge factor and model performance can't be looked at in isolation without that.

>>106168998
>https://simple-bench.com/try-yourself
How many times are retards going to fall for the """intelligence benchmark""" scam?

>>106170794
How long until then? About two weeks?