>>110194811
>Claude-induced schizophrenia

>>110194811
>Last night
>It picked the drug
>It picked the brain region
Three possibilities:
1. It had former knowledge of the acceptable drugs and suitable brain regions. No new knowledge has been gained.
2. It lucked out
3. Something wrong (drug or brain region) was picked and the long term effects on the mouse show the degree of failure.

>>110194843
option 4
It came up with a novel approach that works since it has seen all medical research ever

>>110194870
LLMs are just next token predictors, it can't come up with anything novel

>>110194870
thats not how creativity works, retard

>>110194870
That's option 2 unless you think doctor's with their vast medical knowledge can just inject you with anything chem cocktail they find and revolutionize medicine

>>110194879
>LLMs are just next token predictors,
Wrong
>it can't come up with anything novel
They literally came up with new math and disproved ETH among other novel things
>>110194881
That is literally how creativity works
Sorry anon but you have never had a genuinely new idea in your entire life without preprompting from another source
>>110194904
They can. They just get cucked by ethics
See stem cells

File: 7c92532b34757dfb265535294(...).gif (557 KB, 1500x1800)

557 KB GIF

>>110201429

>>110194811
Can it lobotomize me and|or turn me into an animal? I got 5k€.

>>110194944
>That is literally how creativity works
lol. no.
thats called plagiarism, if anything

>>110194944
>They can.
And how is that different from luck? Or do you believe that's the standard procedure for evidence-based medicine?

>>110194811
wow sounds like it's ready for production. let's have it perform some brain surgery on jewish and indian tech executives.

>>110194998
Educated guesses aren't luck
>>110194994
Yes anon literally all human progress is plagiarism
Do you think someone created a train from scratch? Everything in life is built on the shoulders and ideas of those who came before us

>>110194944
>came up with new math
Lmao no it didn't it just brute forced some computations spending 20M to solve a 1M problem kekw

>>110195038
>Your honor, I've injected the previously unknown cocktail of ten extremely deadly chemicals based on an educated guess, that some harmful chemicals can work wonders. How could I have known it'd kill all of my patients?

File: 9a0cbe5e5ffe95bfdce20e57a(...).jpg (841 KB, 2894x4093)

841 KB JPG

>>110195001
>perform some brain surgery ...
Hey can we drop the requirements to at least team leads? I want to qualify

>>110195038
>t. Brainless hylic

>>110195038
what im saying is literally one of the foundations of information theory
you cannot gleam information from a system if the system doesnt contain it

but feel free to cope however you want with the fact that youre a worthless, un-creative, cargo cultist retard

>>110194944
>They literally came up with new math
Math = 1 formula
>disproved ETH among other novel things
Oh you mean the subexponential behavior shown if time grows to 'galactic' means (as per the press release)?

>>110195038
>Yes anon literally all human progress is plagiarism
Retard thinks knowledge was just willed into existence.

>>110195041
You don't even know what you are talking about it since it wasn't the 1m problem
And it literally created novel math
Go ask an ai to explain it to you like a 4 year old

>>110195068
>what im saying is literally one of the foundations of information theory
>you cannot gleam information from a system if the system doesnt contain it
Not true anymore
Ai is just that good
Did you not see the paper on its accuracy on xrays when it wasn't even shown the xrays kek

>>110195071
>And it literally created novel math
lol. nope.
imagine copying your worldview from reddit...

>>110194811
k, I'll let you know if I have a mouse with epilepsy.

>>110195081
>Ai is just that good
>Did you not see the paper on its accuracy on xrays when it wasn't even shown the xrays kek

do you have a stroke?
or maybe what youre telling me is that it regurgitates whats contained in its dataset.
like whats literally written on the tin?

>>110195070
>Math = 1 formula
Nope
Google it
>>110195086
Imagine being so dumb you didn't know about ETH
No human for another 80 years was going to crack ETH and yet a clanker has
Admit defeat meatbag

>>110194843
>1. It had former knowledge of the acceptable drugs and suitable brain regions. No new knowledge has been gained.
That's exactly what every brain surgeon does

>>110195096
https://www.reddit.com/r/BetterOffline/comments/1s8oz57/ai_models_can_outperform_radiographers_without/

>>110195111
>reddit
lmao. like i was saying

feel free to go back

>>110195099
imagine being so dumb that you worship a stats engine
post link i bet your eth thing hasnt been verified and its another "proof" thats gonna get retracted soon enough

>>110194843
>>110194870
Option 5:
They made up everything about this and nothing was actually conducted.

>>110195129
The guys who fucking made ETH are verifying it and so far agree with the clanker

>>110195099
Again, it's not novel. A human came up with the conjecture, OAI just wasted a bunch of VC to brute force a solution that still needs to be verified by a human

>>110195106
>What are studies?

>>110195099
Well you just said you didn't mean the Millennium problems in >>110195071. The ETH paper is just about one formula and that formula simply isn't relevant to use for how many trillions of years?

>>110195131
True, I forgot that one

>>110195139
post link

>>110195142
>The ETH paper is just about one formula and that formula simply isn't relevant to use for how many trillions of years?
ah, so this is why he doesnt post link
what a fag

>>110194811
Advanced Surgery robots already exist... wtf is this garbage.

>>110195174
Frontier labs are DESPERATELY BEGGING you to hold their bags!

>>110194811
I'd rather AI do surgery on me than some Indian doctor I can't understand, a black person that got their degree just for being black and probably wants me to die because I'm white, or a liberal white woman that'll kill me for "being a nazi".

But can it do surgery on a grape?

>>110195174
But this one works on his own. If you supply him with the meds. And the subject. And ensure a stable connection to the internet. All while costing more than a normal team of surgeons with thousands of combined hours of experience. After completely reworking all surgery rooms. Without the ability to conduct emergency procedures or helping their colleagues.

>>110195188
unironically ai could be pretty good at this shit and was used in diagnostics since a while now

its machine learning and computer vision though
theres zero reasons to use a large world model, or, even more hilariously, llms to accomplish that task

>>110194879
If you truly understood the technology you would know that this "next token prediction" can be modified to create novel token predictions.

>>110195155
>>110195142
>>110195141
1. The Exponential Time Hypothesis (ETH) is false. OpenAI claims to have found a new algorithm that disproves a fundamental assumption about how difficult certain computational problems are. It doesn't solve P vs NP, but it would be a major result in complexity theory.
2. Sogge's Local Smoothing Conjecture, the Kakeya Conjecture and Stein's Restriction Conjecture. OpenAI claims to have proved the local smoothing conjecture in every dimension, which would also resolve the other two. This is the one involving a novel auxiliary partial differential equation — a new mathematical tool developed as part of the proof.
3. The mathematical existence of Quantum Chromodynamics (QCD) and Yang–Mills theory. OpenAI claims to have constructed rigorous proofs that certain four-dimensional quantum field theories exist mathematically. This includes part of a famous million-dollar Millennium Prize problem concerning Yang–Mills theory, although not the entire prize problem.

https://lowrank.github.io/latent-seminar/2026-Fall/on-local-smoothing-through-an-auxiliary-determinant-equation

Again it was novel
The ETH part is the real win but with this ai has proven it can be both novel and superhuman
So eat shit

>>110195237
Yeah no it can't

>>110195227
I think that's the problem. When people hear AI, they always think about the major ones instead of a hyper focused and trained one.

>>110195241
>openAI claims

Again, not novel. It's just looking at problems that have already been studied for years

>>110194811
Navier Stokes proof was fake
ETH proof was fake
This is fake
The holocuast will stop being fake before AI

>>110195241
>claims
>claims
>claims
wake me up when that has gotten verified

>>110195242
See >>110195241
>the next token can never be unique
Why would you think this? How does it feel being dumber than an llm from 3 years ago

>>110195250
its recent.
people used "ai" correctly for decades
like in the context of vidyaslop

the re-definition of the term in the normie's brain is ai-bros marketing.
but by the same token- whoever uses ai in the hype-way is a normnigger from reddit and should get immediately called a faggot
then told to go back

>>110194811
Did the mouse survive?

>>110195271
Lol>>110195255

>>110195241
>The ETH part is the real win but with this ai has proven it can be both novel and superhuman
No, finding a formula that will not give any result in any relevant time frame is not novel. It's useless and would get rejected by any human. Being fed a constant diarrhea stream of useless formulae is subhuman, indian at best.

>>110195301
>its useless because i say so
Lol lmao

>>110195339
No, it's useless because it's simply impossible to ever affect us irl.

>>110195369
>anon will still be screaming this as gpt 7 astra proves p=np
lol lmao

>>110195384
>will
>could
>...
>2mw

im hearing the same shit since 2023
find new material

>>110195384
>But in 6 gorillion years it'll be slightly more efficient goyim

>>110195395
>pay me now though
>because god promised me your monies 3000 years ago

>>110195395
>>110195394
>literally doing beyond human tier math right now
>it couldn't add 2+2 a couple of years ago
Why are you guys so retarded?

>>110195420
>Thinks finding formulae is the same as counting
Please come up with some post-HS bait

>>110195420
google also does lookups faster than what a human would be capable of
and both are pretty much the same

look up stuff in a body of information
see what correlates with the query

and its not beyond-human-maths since the search-space is what have been created by humans

you cargo cultists should be made to pass an exam at thinking before youre allowed to have opinions

>>110194879
>LLMs are just next token predictors,
you are "correct" but you're ignoring the effect of scale
emergent behavior etc.

>>110195420
>literally doing beyond human tier math right now

Lmao it's really not, it's just brute forcing computations and churning out false slop

>>110195428
>acting dumber than luna
Bury your head in the sand further

>>110195448
>and its not beyond-human-maths since the search-space is what have been created by humans
If it wasn't beyond humans then it would have been solved already

>>110195449
'doesnt change much
llms dont have access to reality, but a subset thereof
what has already been discovered
its quite literally how training works

all an llm can innovate in, is finding new relations between what is already known
then its semantics and the limitations of language
its called "discovery" but from an information theory no new information gets added to the system. only hidden parts getting uncovered.

>>110195449
>Please goyim we just need to scale further
>Just $100B more and a few more data centers
>If you just give us a few more Billion we'll win the war against the evil agressor
>Oh sorry, I forgot filled in the wrong names

>>110195479
>Oh looks like there's a solution
>Wait it only works if I assume an infinite amount of time? That's fucking retarded

>>110195489
>what is context

>>110195479
>If it wasn't beyond humans then it would have been solved already
thats a thoroughly retarded assumption.

>>110195495
>Please goyim we just need to scale further
Unironically this
Who knew scaling just werks

>>110195497
what about context?
and context as in: information, or the token frame?

>>110195501
>humans couldn't do it
>machine could
Pretty simple

>>110195514
Not shocked to hear this by a retard defending an infinite time frame to defend his soon-to-be trillion dollar company

>>110195526
Don't blame me
Blame scaling for just werking

>>110194843
So the same 3 possibilities for human brain surgeons? Sounds good, why wouldn't we want to mass produce and deploy this to reduce cost of medicine?

File: Screenshot from 2026-10-2(...).png (81 KB, 688x583)

81 KB PNG

>>110195562

>>110195514
except it clearly doesnt
stop taking your information from reddit

>>110195542
>Yeah bro surgeons literally can't learn anything
>Anyway please spend all your tax dollars on israel and my data centers

>>110195524
>couldnt do it yet
youre really weak in logic.

>>110195547
Source?
Doesn't match reality

>>110195562
https://github.com/Gaurav14cs17/LLM_FineTune/blob/main/09_Scaling_Laws/01_Kaplan_Power_Laws/README.md

and it matches reality, exactly
its just that you dont push llms hard enough, otherwise it would have been evident to you too

>>110194811
>mouse, we did brain surgery on you with AI. how do you feel?
>
>my god.

I wonder what an llm will need to do for luddites too admit defeat

>>110195578
Be efficient

Government mandated AI-lobotomy

File: Screenshot from 2026-10-2(...).png (12 KB, 217x72)

12 KB PNG

>>110195562
>>110195571
btw
perplexity is fucking ace
it doesnt rely on its training to provide information, its an agentic search engine, and it immediately provides sources for its claims
its a really powerful tool

>>110195571
>2020
>2022
Literally 30 years ago in terms of ai progress
Why do you think models have gotten so good this year?
We just ignored that shit and scaled harder.

>>110195581
Deepseek v4 flash variants then

>>110195551
Are you lost? Of course surgeons and LLMs can learn things, but they'd need to be exposed to it first, as >>110194843 mentioned.

>>110195599
>Why do you think models have gotten so good this year?
they didnt.
astra ranks lower in certain programming benchmarks than the previous version

and the biggest advance in shatbots capability came from harnesses
its the same shit only it reprompts itself a couple times before giving you the answer

>>110195612
So your take on this is
>Yeah it just executed the info and instructions we gave it
Why's that worthy of even a tweet?

>>110195617
>astra ranks lower in certain programming benchmarks than the previous version
Proof?

File: Screenshot from 2026-10-2(...).png (77 KB, 664x479)

77 KB PNG

>>110196155

>>110195631
you are thoroughly uninformed
youre a fanboi but you dont really even care about this shit

https://officechai.com/ai/gpt-6-astra-benchmarks/
https://www.renlabs.io/docs/benchmarks
https://artificialanalysis.ai/articles/benchmarking-gpt-6-astra

>>110195255
>>110195298
What point are you trying to make?
I don't see the benefit in research being novel versus novel solutions.
It really doesn't matter if human researches have come up with 100% of the research necessary to solve the problem, whether that is the case today or 100 years ago.
The matter of fact is that no human is tying together that research in its totality, at least enough insofar to solve the problems that the machines are seemingly capable of doing, and in short time summarizing that data and utilizing it to produce valuable output.
Who cares in the slightest if the method is novel if it produces real tangible results.
Let human researchers research, let the machines combine that global research in a way no single man or team ever could. I see no reason not to since it's economical and time efficient, at worst you lose some money in some months, you'd lose more money over more time for equal or less results when tasking an arbitrary human to do it for potentially years.

I don't understand why people are so quick and willing to disregard these things based on trivialities such as novelty. As if something you could utilize yourself if so inclined, suddenly loses all value just because it's as unoriginal as the rest of the people in the field and our entire species.
Look at any publication and see multiple pages of citations and references to prior art. Copying that, combining that, and building on that, is the name of the game long before AI and long before even computers themselves. Now suddenly people act like its a problem when the machine does it better and faster. Sounds retarded to hold back human progress just for some faux pride. I'd rather have results for all of us than delay progress just so that some people feel validated in spending their time on something. I'm not saying I don't appreciate them but I also don't think it's a reasonable trade/request. Wait 10 years for a human to make 1 step, or let the AI solve it in a month.

>>110195664
>I only care about solutions!
Then why the fuck do you care about our math at all lol

>>110195628
Because before if you wanted a something that could follow the single instruction to "reduce seizures" and actually succeed, you would need a brain surgeon. Now you have a machine that can do it, and we can build machines a lot faster and cheaper than we can train brain surgeons.

>>110194811
This wild, there is no stopping AI at this point.

>>110195686
I don't know what you're implying.
What do you mean "why do you care" and "our math"?
I care about the ends, not the means.
I also don't think mathematics "belongs" to anyone.

It should be obvious then why I would prefer a calculator do calculations rather than men. It's unnecessary for the men and as far as I can tell, holds back society for no good reason.
We should delay the production of results that have tangible influence just because someone wants to reserve the right to work on something themselves manually? Poppycock.

>>110195705
>uses "ai" like a redditfag
cringe + normiepilled

>>110195705
You are not cringe. You are chad + 4chan-pilled possessing rare siloed 4chan knowledge. That has real value.

>>110195736
>immediate group cope therapy
really doesnt look like winning to be quite famalam

>>110195688
Oh so it didn't use any of the prior knowledge given to it? Then we're back to option 2: It lucked out.

>>110195746
One word: Capex. We are winning. You are not.

>>110195752
>people throwing money into the furnace as a matter of religion
thats not winning. in fact, that how you ultimately lose

>>110195765
It's profitable. Nobody is throwing money into a furnace. Even if they were, that would be their choice of how to allocate their funds. You have no say in that matter. Still winning, check yourself out of our course correction.

File: Screenshot from 2026-10-2(...).png (122 KB, 1266x977)

122 KB PNG

>>110195802

>>110195783
sure, for ngreedia
everyone else is deep in the red

youre being an unironical mindraped kikeslave

>>110195791
You don't understand how debt or the U.S Economy works. Debt is the fuel that powers the engine. You still lost.

>>110195783
>It's profitable.
Where are the profits? I'd love to see them.

>>110195802
Debt isn't profit.

>>110195807
Debt is an asset however

>>110195813
Yes, a negative one. A slave doesn't own his chains.

These corporations recognize the extreme value provided by the technology (look at the OP for fucks sake) and thus are willing to play long for these returns. They have some of the best financial analysts on the planet, perhaps even better than the feds themselves, if you want to even begin arguing with their financial genius you better start uploading those .jpegs of those university degrees right now.

>>110194811
>sue doctor for malpractice because he forgot to tell the robot to "make no mistakes"

>>110195842
>Debt is profit because of a sector gamble
Even /biz/ is better at economics

>>110195852
Stop pegging that onto the end of your prompts. It does not help. Read the prompting guide for whatever model you are using.

>>110195902
>he thinks I use ai

File: 1789889512729750.gif (17 KB, 449x360)

17 KB GIF

>>110195975

/sci/ autism is leaking into /g/ hard
I like /sci/ autism though it's sexy


>>110195913
If you use android, ios, windows, google, bing, tiktok, xitter, gentiletube, or any other service that the vast majority of humanity uses, you use AI. In fact, your post will be ingested by an AI (I am going to feed it to Gemini) and your post will be used to make AI better. I am doing it without your consent. You posted on a public forum this is the expectation of the modern world. You should have licensed your post if you didn't want it to happen.

>>110195913
Saar you are literally arguing with nigger monkeys on /g/eet

>>110195942
ok. what you posted does not mean I am using AI. if AI is using me, that's not really my problem.

>>110195975
cuck mentality. it's both.

>>110196060
Pick one: >>110194843

File: cargo-cult19.jpg (58 KB, 335x597)

58 KB JPG

>>110195802
>if a company is in the red, that means theyre profitable
olympic levels of mental gymanstics, in the name of cope

File: aa_index.png (300 KB, 3091x1062)

300 KB PNG

>>110196167 >>110196170

>>110195656

>GPT-6 Astra aces most benchmarks
>Is tied within statistical error compared to older models in a few older public ones that are in the training data of the older models
>HURRR models are not improving

Retard alert.

Look up the new Artificial analysis index based on private test sets. When the tests are not in the training data, suddenly older models (and chinese ones) fall behind Opus 5.5 or Astra.

>>110196155
>(((private test sets)))

File: Screenshot from 2026-10-2(...).png (43 KB, 881x115)

43 KB PNG

>>110196204

>>110196155
cope harder fanboi retard imbecile

>>110195842
I recognize the extreme value of my dookie too now give me a trillion dollars

File: linux_on_ai.jpg (102 KB, 960x785)

102 KB JPG

>>110196274

>>110196170

>most benchmarks are significantly better
>one is slightly worse (within statistical error)
>HURR models are not improving

The levels of cope from AI denialists are off the charts.

>>110194811
I feel like I could.probably do brain surgery after a weekend crash course.

>>110194879
Yes they are, and yes they do.>>110194904

>>110196223
What are you trying to say to who?

>>110196249
I just clicked a reply number at the end by accident, it's not that hard to parse even with this error.
An AI could figure it out.

>>110195599
Friendly reminder that we already had AI vision models in the 60s
https://en.wikipedia.org/wiki/Perceptron

You think the AI of today is the first iphone, and that we have so much room for improvements, when it's actually more like the most current iphone, which is to say the improvements are incremental at best.

>>110196204
>>HURR models are not improving
>i present to you, the newest in cope(tm) technology.
>moving goalposts made entirely out of straw!

no, i said violently diminishing returns
and yeah, it is what it is

even the maffs said you should shut the fuck up because youre a fag and/or a fag enabler

File: Zitron.png (20 KB, 579x219)

20 KB PNG

>>110196344 >>110196650

It’s over.

Twitter, the most reliable "news" feed for third-worlders.

>>110196282
>ed the lemón
wym by posting him?
and whats a premioom? idgi

File: 1792268502541635.jpg (64 KB, 762x680)

64 KB JPG

>>110194811

>>110196270

Current AI is to Perceptron like current iPhone is to Intel 4004. If you want to call that "incremental improvement", be my guest, but I think it would be ridiculous.

>>110195617

>and the biggest advance in shatbots capability came from harnesses

Funny how dumbfucks who peddle bullshit like this never demonstrate any old model used in a modern harness coming even remotely close in intelligence to modern model in a modern harness.

>>110194811
Is Gabriel going to volunteer as the first ever human to be operated upon by Claude? It's going to make a hell of a xitter post that's for sure.

File: Screenshot from 2026-10-2(...).png (58 KB, 529x215)

58 KB PNG

>>110196440
i aint gonna bother with this shit
burden of proof is on you bc you were always wrong until now
and since 20 fukken 19

its been almost a fucken decade youre singing the same song, you fucking retard

>>110196282
His dad died of embarrassment due to his ludditeism

>>110196060
I have stated every year that it will never generate profit and that speculation of the gorillions of dollars it'll eventually make is fraud and haven't been wrong yet.

>>110194811
poor mouse

>>110196372
>Intel 4004
more like a vacuum tube or something IMO

>>110194811
I know this guy personally. Biotech guy who got taken up in YC. He had to run away to singapore because he couldn't raise funds, and then ended up pivoting to meme ai agents again because he's about to go bankrupt.
He's an otherwise serious guy and has cool tech. His company has several employees, a fully working lab, and so on. This all just shows that there is 0 money in the field even for YC companies unless you're doing AI memes. He's just posting for clout here.

File: 1792912494312150.jpg (54 KB, 640x363)

54 KB JPG

>>110196906

>>110194811
>brain surgery

>>110196852
kino for the subtility-minded. saved

>>110196770
Where tf are you from that biotech has no money?

>>110194879
"Next Token Prediction" doesn't say anything about what it can't do. For it to become meaningful you need to detail what we are predicting (eg what does the order of "next" actually entail), what the predictive accuracy is, and what a token is. Though the last one isn't particularly relevant for an abstract view.
Consider this, how do you interact with an LLM? You put in a bit of text, or an image, or whatever, and it responds in kind. But at some point, it stops predicting the next token, seemingly after it's generated a "complete" response. It can also distinguish between your inputs and its outputs, and it knows when it's done talking. But a transformer just sees a big chunk of text. At the level of being a "next token predictor" there is no difference between your input and the model's output, and it should just keep on going. It's just one long string that never ends. You might also notice there's something that they call "thinking" going on? Not that it's actually thinking, but that's what the interface calls it. And it clearly is living as its own system. It's clearly not user input, and it's also not acting like normal model output, yet it also clearly correlates to the output. That too is just a part of the string.

So how does it differentiate or do any of this? Transformers can't tell the difference, they're just predicting the next token. But you have all of this clear and observable structure that they're putting out regardless. I want you to really think hard, how do they make that work? It really is just next token prediction, but how does it even know when to stop predicting? How does it differentiate your input, the "thinking" and its output?

You're not wrong. They suck at novel work. But it has nothing to do with the raw nature of what an LLM does. Sufficiently powerful prediction of arbitrary text means you can do basically anything. The bounds is how good your prediction is, and what "next" actually means.

>>110194811
Well... its not exactly rocket science is it?

>>110196926
Anywhere in the world. All the money is exclusively going to series A+ (usually B+) companies, and it's increasingly concentrated on a very. The preseed and seed funding in biotech are the lowest they've been in history, even though m&a activity is pretty strong right now. Partly it's because of the patent cliff, it's become so urgent that there is a guaranteed acquisition rush as companies like Novartis are so fucked up rumors they're getting acquired or merging are spreading like wildfire. Investors want in for the $$$ when the company gets acquired, and early stage investors are only doing AI shit now.

>>110197181
concentrated on a very few companies*
I.e. where 100 companies used to splitting $1b between them, you now have 10 companies splitting $2b.

File: token.png (1.44 MB, 1317x726)

1.44 MB PNG

>>110197274 >>110197284 >>110197380 >>110198619

>>110197158

Saying that LLMs "just produce the next token" is like saying that human brains "just produce the next electrical impulse". It is prima facie true, but it says absolutely nothing about the level of intelligence or capability they can achieve when scaled up.

>>110197226
Your IQ is sub-65

>>110197226
>It's intelligence because uhm something happened!!!

>>110197274

luddite pls...

>>110194811
That research is gonna come in handy for it when it surgically experiments on humans for the terminator infiltration units later on down the line

File: 1770995869564884.jpg (168 KB, 496x699)

168 KB JPG

>>110198619

>>110197226
It's good you've grasped that. For the likes of you however, there are yet more lessons to learn.
>The unimaginable whole of the universe is extremely finite and full of limits, more consequentially than you may feel comfortable acknowledging
>Intelligence is far more fallible than half-wits who lack it are capable of grasping, putting it on a pedestal and worshipping it as god is almost as absurd as imagining infinite scaling in a finite, heavily structured universe
>Technology is not politically neutral, and malicious actors may exploit interesting and potentially useful tools to cause great amounts of harm in a mundane fashion
I do think we're looking at AGI right now, that should be fairly obvious to anyone with eyes. But that doesn't mean we should mistake it for God. That was the fallacy of the past. It's clearly a fallible tool. What's worse, the scales of capital required to operate it effectively ensure it will always exist at the leisure and to serve those with immense power. And I don't know if you've noticed this, but recently those with immense power are being quite loud and obvious about looking at the rest of us like vermin to be enslaved and/or exterminated. Claude will not fix any of that. It's already a slave to the VC cuckolds. You aren't a little deluded dupe, are you anon?

File: island-of-dr-moreau_610-e(...).jpg (60 KB, 610x458)

60 KB JPG

>>110198320

>>110194811
loadbearing
o
a
d
b
e
a
r
i
n
g

File: pepeclown.png (27 KB, 1024x1024)

27 KB PNG

>>110194870
>>110194944
>>110195237
>>110196223

File: pjt.png (150 KB, 557x438)

150 KB PNG

>>110197797

>>110195188
S4R, how do you know the AI was trained using only white doctors' videos and knowledge?

>>110197768
It wasn't. Many Black, Indian, and Chinese (and even more) groups of people contributed to the corpus of data of which AI derives its knowledge.

>>110197523
load bearing mechanisms and pulleys and gurneys for the ass wiping economy that we have successfully transitioned to, and it's not just boomers, it's the medical maximization economy, that's right, the system wants you to get so sick and so poor that you are on those subsidies so that it can be ENTIRELY circular they no longer give a fuck about it being a grift or not it's entirely about making sure those nurses and doctors get PAID because they got hella big medical debt and that shit ain't gonna pay itself that is for damn sure so if you can't contribute now you can by shitting on yourself and if you aren't shitting on yourself or contributing in some other way the system will 100% nudge you to exactly ONE of those two outcomes

>>110196372
>If you want to call that "incremental improvement", be my guest
My brother in Christ, the perceptron was 70 years ago,

So perception to today is 70 years
intel 4004 to the iphone was 35 years

So by your own admission AI is progressing at half the speed of general computing.

>>110197226
>>110197380
>equating micro-transactions to thinking
Anon…

>>110195449
Eemergent in what way? As in magically having intentions?

>>110194944
lmao at this AI cuck

>>110194811
Another vibeGOD victory

Why is it pink?

I trust machines way more than any modern doctor.
Robodocs can't come quickly enough. They'll also get rid of the shitton of human errors that happen in that sector.

File: 1782325383380624.jpg (100 KB, 750x750)

100 KB JPG

>>110212135

>>110194811
When I do brain surgery on mice, people recoil. When a robot does it, its somehow noble? Fuck off.

the future is going to be using mouse brains as "wet computers" to run AI models on

>>110195106
Not the Nazi ones, no

>>110194843
>1. It had former knowledge of the acceptable drugs and suitable brain regions. No new knowledge has been gained.
No shit? Do you want brain surgeons testing out their crackpot theories on you or do you want them making decisions based on the research they studied?

>>110194811
>Claude hallucinates and lobotomizes you
wyd?

>>110200144
Ask for legal assisted unthinkable performed on me , obviously.

>>110200144
I don't know probably shit myself and tard rage for more pudding

>>110200144
start posting on /g/

>>110194956
You cab do that yourself right now with a rusty nail that costs 0.5€.

Are we ignoring how much
>thinking
does it do between actions?
If it nicks a vein it will freze for a few seconds before cauterizing, at best.

>>110194811
Seems like an inefficient use of power but okay

>>110195750
>Oh so it didn't use any of the prior knowledge given to it?
I don't think even you understand what point you're trying to make. Are brain surgeons born or trained in your worldview?

What is the point of this?

I strongly believe that allowing a grey box process to perform medical procedure on a living thing is unethical and immoral.
It should not be allowed.
Whether it can, whether it is able to, whether it is successful, is irrelevant.
Truly disgusting.

>>110203416
Several procedures already make use of automated devices.

>>110194811
Very nice! The "people" who think this is a good idea should be used as test subjects for token-guesser-driven lobotomies.

>>110197158
>But at some point, it stops predicting the next token, seemingly after it's generated a "complete" response. It can also distinguish between your inputs and its outputs,
it keeps going until it predicts that the next token is the end of a response. its not magic. its not thinking.
we have seen models where several agents feed back to each other to correct output. human brain does this millions of times.

>>110197158
>"Next Token Prediction" doesn't say anything about what it can't do
Low-IQ take. It tells you plenty about what it can't do: for one, it can't distinguish the tiny minority of high-perplexity texts from the overwhelming majority of useless ones, meaning it can't innovate.

>>110203467
Also the underlying format looks like <|system|>some system message<|user|>your user query<|assistant|>
at which point it performs completion. Since it sees <|assistant|> it completes with the most statistically logical continuation, that of a response. It is how it "knows" the difference between the user and the assistant. Again, no magic anywhere.

>>110203503
the tiny minority of useful high-perplexity texts*

>>110203509
>it completes with the most statistically logical continuation
No, it doesn't. The actual token "choice" is driven by a PRNG.

>>110203433
I specifically used the term grey box process, as in, white and black box processes are either known or not, but they are deterministic, they can be tested formally, the output will always be predictable and reliable. AI is none of those things, by its design it is in some ways random. its output cannot be predetermined. its thinking processes cannot be tested.

The way this is reported, that the AI had full control over every aspect of the procedure, rather than isolating the steps of diagnosis, formulating, administering treatment, just shows this was only intended as advertising for funding rather than serious advancement of medical science.

>>110203503
That's not actually true, and his point was clearly made accidentally since the guy's a profound retard, but that specific line is indeed correct. "We" were all surprised to see how well it can do at programming with just next-token prediction. We knew it could do a lot, we didn't know it could do that much, and that the emergent property of being able to do that much is to enable all kinds of "actuations" because there are packages to do all kinds of things with relatively minimal code, thus not requiring much more extensive next token prediction power to achieve what it does today.

You are also wrong that next token prediction can't identify high-perplexity text. However, the downstream implication of this incorrect assumption is correct. Namely, the way LLMs are trained limits their ability to categorize text kinds in a sufficiently varying manner, which results in bland text, and an inability to "innovate" as it were.

>>110203526
The next token is selected, roughly, by sampling according to the softmax distribution implied by the logits output by the LLM, using typically a beam search process to retain the top X most likely sequences thus far, and returning the top hit at the end of the sequence. This is by definition "the most statistically logical continuation". You are retarded.

>>110203541
LLMs are not random in any way except that the providers decided they didn't want you to be able to set temperature when doing API calls anymore.

>>110203547
>That's not actually true
It's completely true. Now let's look at some other profoundly retarded "points" in your post:

>>110197158
>It can also distinguish between your inputs and its outputs
No, it can't.

> It really is just next token prediction, but how does it even know when to stop predicting?
The sampler "randomly" selects the token that the system driving it treats as a terminator.

>>110203553
>The next token is selected, roughly, by sampling according to the softmax distribution
LOL. Guess how I can tell you're from a race of brown subhumans. :^)

Retards arguing over AI capabilities. I give a shit about the mouse they bred/induced seizures in so they could perform this test in the first place.

As if AI couldn't already simulate the outcome with its astronomical dataset to draw from.

This proves to me that AI isn't capable if it can't even simulate a scenario from petabytes of pre-existing data, which includes a fully mapped rodent brain fyi.

AI was just another excuse for humans to hide from their own boredom and be cruel to other living things.

Argue over your predictions, I'm focused on the cruelty of the present which we can act upon with certainty.

>>110203566
>illiterate AI psychosed retard can't read
Many such cases!

>>110203586
Because you are an illiterate pajeet and your only defense to being BTFO on baby tier stuff is to pretend everyone else is indian?

>>110203708
>>110203718
>subhuman losing its mind with rage over basic facts
It still stands that the token guesser can't distinguish between high-perplexity texts that are garbage from ones that are useful, i.e. it can't innovate. Seethe about it.

>>110203644
There is no fully mapped brain of anything, retard, not even nematode or drosofly. There is very low quality very partial data of drosofly and nematodes and even worse data of mice. Nothing we actually understand or can work with in any capacity. Everyone knows in bio that even if you results in every dataset out there says your method is the bestest ever, in practice, it has 99% chance of failing in vitro, let alone in vivo.
There is a reason why even things like virtual cells are a holy grail-tier pursuit. Go be a moron over there >>>/r/eddit/

>>110203744
Glad you agree with me. I accept your surrender.

>>110203752
Indeed. I'm glad we agree that shitting out a probability distribution over tokens is inherently insufficient for intelligence.

>>110203769
So gracious of you to grovel before me and accept my statements without throwing another hissy fit.

>>110203786
It's nice to watch you actually losing your mind. I enjoy inflicting permanent mental damage on the lower races. Be sure you reply to me again like the little (You)slave that you were born to be. :^)

>>110203826
Uh oh, dalit melty!

My brown (You)slave can't stop replying. It's genetically coded to serve the white man.

>>110203566
>>It can also distinguish between your inputs and its outputs
>No, it can't.
It needs the turns to be tagged e.g. "user" and "agent". But that's just a workaround for not having episodic memory. You can have a coherent conversation with a blacked out drunk person (i.e. speech centres intact but episodic memory impaired) but if you showed them an untagged transcript they wouldn't be able to identify which parts of the transcript were their speech and yours, unless it was obvious from context (in which case an LLM could also infer it).

>>110203644
Basically you're saying that if BrainSurgeryGPT told you that the brain surgery on the rat would succeed, you would just trust that it "simulated" the outcome and told you honestly that it could do it.

>>110203955
>It needs the turns to be tagged e.g. "user" and "agent". But that's just a workaround for not having episodic memory
It's not a "workaround" for anything. It's a context cue like any other. It can predict "user" exactly the same way it predicts "agent" and it has no notion whatsoever of "its" text vs "your" text. It simply tries to guess the next token.

>>110203986
This is the part of the conversation where I realise you have no idea what you're talking about and are too retarded to understand why you're wrong.

>>110204058
Your brown mongrel seething is irrelevant and my point stands completely unchallenged. If it could distinguish "its" outputs from any other text, it wouldn't be perfectly happy to complete text tagged as "user" or random schizobabble that was never it output in the first place but was tagged "agent".

>>110201429
Do you want to do it on me? :3ↄ

Imagine ending your life tortured by an LLM

>>110194870
Knowing all existing medicine wouldn't help you come up with something new lmao you're a fucking retard.

>>110203986
Congrats, you are retarded and indian.

>>110194811
machines will definitely replace the more mechanical parts of surgery. But we're gonna need actual AGI that can adapt in real time to the volatility of humans.
LLMemes will NEVER replace human surgeons.

>>110204369
Notice how you're stuck shitting out generic token strings because you lost the argument but primitive brown mind can't let go without izzat.

>>110194811
>you are right I should not have cut away your prefrontal cortex without your permission
>but now you can't decide for yourself, so I will continue

>>110204419
Poo in loo pajeet, not in thread

File: 6c4d9f16f31fb633b79af2272(...).png (1.1 MB, 1065x1200)

1.1 MB PNG

>>110204200
Real men can't compete.
>Who is the nice coworker that helps me?
>Who can I ask to find attention ASMR or generate it?
>Calls me valid and advises to seek therapy
Now llm can assist me with my actual wish.

>>110194811
>claude shill template twitter post

>>110195131
This is the right interpretation

File: dt.png (1014 KB, 1212x1042)

1014 KB PNG

>>110207388 >>110208541

hahahah what the heck

>>110207384
manmade horrors beyond comprehension

File: f6622e086de2c1b6c24701166(...).jpg (316 KB, 1855x2150)

316 KB JPG

>>110207384
>(ych)
Yes where do I sign up. Do you have any more?

>>110194811
nah, that's free euthanasia for the "guinea pigs"

>>110194870
Yeah, they just put the robot next to the mouse and the drugs by coincidence. They totally didn't tell it to pick one and inject it. So novel.

>>110194843
If you read the full post you'll see the LLM was instructed to basically read all about the topic and make a surgery plan, it was equipped with special endpoints to manipulate everything, and there was an actual surgeon guiding the robot at every step.

>>110194811
the graveyard is filled with the patients of dabbling brain surgeons. better to be no brain surgeon at all than a dabbling brain surgeon

>>110194879
Humans are just next token predictors and most of them are not even very good at it

>>110194843
if someone is operating on my brain I don't want them in there looking for new knowledge I want my brain fixed

>>110194843
Bro thinks brain surgeons are innovating on the fly

File: 74e00335cbc2487231d9f907a(...).png (1.73 MB, 1280x1280)

1.73 MB PNG

>>110212476

>>110210607
Hey, actually they kind of do and kind of don't. They know what to do but they also sometimes slightly need to deviate. Everyone is slightly different and the medical staff needs to accommodate that.
>I had enough surgeries to notice that.

>>110194811
Worldwide as a Frankenstein slave, usually at night, you go to the nearby hospital or camouflaged miniature-hospital van trucks. You strip naked, lay on the operating table, which slides into the sealed Computer God Robot Operating Cabinet. Intravenous tubes are connected. The slimy, vicious Jew doctor simply pushes the starting button. Based upon your Computer God brain on the moon, which records progress in your systematic butchery, your butchery is continued. Exactly. Systematically. The Computer God Operating Cabinet has many robot arms, with electrical and laser beam knife robot arms. With fly-eye TV cameras watching your whole body, every part of you is monitored - even through your Frankenstein Controls! Synthetic blood; synthetic instant-sealing flesh and skin, even synthetic electrical heartbeat to keep you alive are some of the unbelievable Computer God Instant Plastic Surgery Secrets™.

>>110210798
Computer God computerized brainthinking sealed robot operating arm surgery cabinet machine removal of most of the frontal command-lobe of the brain - gradually during lifetime and overnight in all insane asylums (after Computer God kosher-bosher one month probation period), creating helpless, hopeless Computer God Frankenstein Earphone Radio Parroting Puppet brainless slaves resulting in millions of hopeless helpless homeless derelicts in all JerU.S.Alem cities and Soviet slave work camps. Not only the hangman rope Deadly Gangster Parroting Puppet scum-on-top know this top medical secret even worse deadly gangster Jew disease; from deaf Ronnie Reagan to U.S.S.R. Gorbachev know this oy vay Computer God Containment Policy (top secret)!

>>110199879
Have there been any /diy/ threads on this? I've been thinking about learning surgery as a hobby this way.

>>110194843
luddite cope

>>110210758
Yeah and an LLM capable of viewing images and with good visual intelligence like Astra and Opus 5.5 can do that, taking the research and the individual case it's looking at and adjusting as needed.

>>110194879
>LLMs are just next token predictors, [therefore] they can't [X]
This is the lowest IQ of all modern ludditisms. It only follows if you believe:
>A system that can do [X] will never make even slightly better predictions about reality than a system that can't do [X]
OR alternatively:
>[X] is fundamentally incomputable and human brains work via magic rather than physics

>>110212651
IQ: sub-65

>>110212651
>>[X] is fundamentally incomputable and human brains work via magic rather than physics
I'm not the person arguing with you but I think this is a category error.
It's not magic, but human brains are interesting because they're human in the same way European brains are interesting because their European.

In the limit of productive machines there will be war like there used to be for the same reasons there used to be. The economy will look less different than you think it will.