Why are people so scared of AI I literally don't get it. It's not very "dangerous" unless you act like a moron with it. It's just a useful tool.
It's a glorified something, but tape recorder make me think he's senile.
If LLMs are ai, so are markov chains and in particular autocorrect.
They literally don't understand it's just a text prediction algorithm. The "AI" branding was a colossal mistake, there's nothing intelligent about LLMs and normies hear AI and imagine skynet
>plumbers not understanding AI isn't just (un)supervised learning
Reinforcement learning algorithms have been shown to execute the operator just to fulfill its purpose better
Whoever thinks that AI (especially reinforcement learning AI) isn't inherently intelligent is a stone age religioustard hurrr man special God thank
Reinforcement learning is the exact same shit. Instead of saying "this is what the correct output looks like" you're saying "this is right" and letting the network figure out what you mean. You cannot train the network without a known input and a known output, which means you can't make a NN to do something unless you have a way to iterate over the result millions of times.
The human brain is the result of training on millions of years of evolution running on real physics instead of shitty binary machines, and you think we can simulate intelligence on a few petabytes of Internet-scraped text and images, or a camera hooked to a few thousands of cars?
>You cannot train the network without a known input and a known output,
Why are morons always this sure ?
RL does not necessarely use NN, they are frequently used as a feature extraction or a value function approximator in some technicques
There is no "known input and known output", this is why RL is harder to train respect to [S|U]ML techniques, you are in the same time "estimating the i/o relation" and finding the optimal strategy that optimize the total expected reward
the first of estimation can be done with NN but it's not the core of the RL method of choice technically
>Reinforcement learning algorithms have been shown to execute the operator just to fulfill its purpose better
No, morons programmed a dangerous machine wrong and then sat in the blast radius. It's no different than shoving the wrong kind of munition into a gun it's not supposed to be in and then having it explode in your face.
>Reinforcement learning algorithms have been shown to execute the operator just to fulfill its purpose better
This is like saying a spider killed it's mate to fulfill it's purpose better because it was programmed to do so and that is bad and all of evolution is bad for creating that behavior because killing is wrong
> Reinforcement learning algorithms have been shown to execute the operator just to fulfill its purpose better
If you knew anything about anything, you’d know that you only need a target function and an optimizer with zero ””learning”” to get there
They literally don't understand it's just a text prediction algorithm. The "AI" branding was a colossal mistake, there's nothing intelligent about LLMs and normies hear AI and imagine skynet
>it's all just varying degrees of undergrad level statistics
if so, then all the things that humans learn can also be called statistical data that is put together to draw conclusions from it
then, what's the difference with AI?
Language models don't really draw conclusions, use logic, or understand what they're being told or what they're saying.
They're kinda like philosophical zombies, now that I think about it.
>it's just a text prediction
ChatGPT is, and no one is fricking claiming it's going to take over the world. It's the implication of its capabilites that is worrying people. You can essentially let it perform self-evaluation as well as interface with other AI-models and applications with some effort. It could theoretically train its own models. The sort of framework that would let it run loose on a large scale in this way doesn't yet exist, but it's going to come eventually. How do you not understand that? Are you fricking moronic or what?
>The sort of framework that would let it run loose on a large scale in this way doesn't yet exist, but it's going to come eventually.
Yeah, yeah, the singularity is nigh.
We don't actually have any clue how long this concept scales. We've used most of the high-quality linguistic training data of the world (could probably squeeze out one or two orders of magnitude, but that's only going to go so far). We've only succeeded in making it worse after initial training, not better.
It's a possible future, but it's far from an inevitability at this point.
Yes, and it would produce garbage. Ml models have no goals or reference points, they just fit the prompt you give them. If you let a LLM train a model you will just get noise.
Go ahead, try to prompt GTP with nonesense, or just a few words. The perceived intelligence in these tools is all user-side, it's like looking at the clouds and seeing the shapes of animals.
And if, somehow, you trained a model to train models that produce coherent outputs you'd just end up with mediocre models. Why? Because ML models cannot produce excellence, by definition what they produce is close to the average. An artificial neural network will never write a novel to rival the classics, will never make a picture to rival anything on the louvre, and will never train a network that exceeds it's own limits. Statistics don't allow such a thing to happen.
>Because ML models cannot produce excellence, by definition what they produce is close to the average.
interestingly, being average in every field is quite above average.
Sure, and that's useful. Will AI take away jobs? Yes, eventually. Code monkeys will be replaced, software engineers won't. Commission artists will be replaced, art directors won't. Technology routinely displaces the lowest rungs of professions.
But being average is not going to lead to any AI apocalypse scenarios. You can't look at a tool that can produce a decently coherent stream of text for a few paragraphs and declare that this technology will turn the human species into paperclips
8 months ago
Anonymous
yeah, i agree with all of that, and until humanity produces a trillion words of higher quality text the cap for LLMs isn't going up.
The reason you don't see it how I see it is because your thinking is locked to the realm of the simple input/output of using text prediction models. ChatGPT cannot "think" by itself nor does it accurately judge what improvement even is or how it is defined. It can however complete tasks that are clearly defined by human input and use external tools if its given instructions on how they opperate, such as when given predefined commands, a manual explanation of what the expected input and output is, and having those read into an external module performing that task. It could outsource all of its operations by the use of commands to such framework built with applications with a predefined goal and perform complex tasks this way.
Nothing about this seems in any way impossible even with current limitations of language models, and it is essentially the foundation of a fully autonomous network that, given the resources, could find flaws, build its own applications, seperate models for operating them and continuously improve though an own internal Q&A. Where exactly does this seem impossible to you?
>It could theoretically train its own models. The sort of framework that would let it run loose on a large scale in this way doesn't yet exist, but it's going to come eventually. How do you not understand that? Are you fricking moronic or what?
it could theoretically provision a billion dollar worth of hardware for weeks, without anyone noticing?
This shit wouldn't be possible without external help and a pre-existing framework to deploy even a basic conceptual system like this. That's why I said it explicitly.
It cannot do it's own anything, it is a slave to the input-output system. It cannot think, it cannot decide, it cannot problem-solve, it cannot even give you a response without being explicitly asked to.
You've skipped hundreds of steps in your reasoning. How are you going to train these auxiliary networks? How are you going to train your main network to use them to complete a task? NNs are trained by labeled examples, thousands to millions of them. If we had a way to train NNs with only a few examples then yes, I agree that might lead to genuine intelligence. But the current "AI" methodologies are completely incapable of doing that and will never achieve anything more that being an input-to-output copying functions
8 months ago
Anonymous
>If we had a way to train NNs with only a few examples
This is of course not actually possible from an information-theoretic point of view.
8 months ago
Anonymous
" If we had a way to train NNs with only a few examples then yes"
This is a perfect example of expecting more out of AI than people in order to decide both are intelligent. How many billions of data points were fed through your eyes and ears and nose and skin before you learned anything. How many millions of years of data being sifted through evolution before we developed the built in hardware to innately recognize an edge and how many small things we can do depend on that.
But AI is somehow suppose to start from scratch and learn everything from a couple of pieces of data and no sensory organs to call its own.
>It could theoretically train its own models. The sort of framework that would let it run loose on a large scale in this way doesn't yet exist, but it's going to come eventually. How do you not understand that? Are you fricking moronic or what?
it could theoretically provision a billion dollar worth of hardware for weeks, without anyone noticing?
You're literally too much of a fricking moron to understand what using external tools means. See my other response. LLM's can't do shit on their own, let alone train their own models, that much is obvious. A fully functional system isn't just a LLM talking to itself.
pretty sure you don't know what you're talking about but i'd love to see you try to explain.
8 months ago
Anonymous
I don't need to explain shit to you: You've already fundamentally misunderstood what I was talking about and imply I think something like ChatGPT can just shit own a better version of itself. You're too fricking daft for this, buddy. Sorry.
8 months ago
Anonymous
i didn't imply anything, i'm not the other anon, but your vague overly-optimistic "tools man tools" shit is hilarious, you've really been sold on the hype.
8 months ago
Anonymous
I'm neither hyped, nor sold on anything. You let LLM's build a chain of operations that need to be performed to complete a task, give instructions on how it communicates to all the relevant applications and define I/O, and then it can do its own thing. HuggingGPT allows you to do exactly that. People either don't understand the capability of this if you scale it up or just never understood anything about it. I'm not going to reiterate the concept for every mouthbreather at this point.
At least the other anon immediately tried to understand the technical limitations. The other moron didn't understand what I was saying at all.
8 months ago
Anonymous
>You let LLM's build a chain of operations that need to be performed to complete a task,
How?
You bring up GPT as an example, but GPT cannot tell you anything it can't infer directly from its dataset. It's literally a more efficient search engine ("efficient" in that you're using a supercomputer instead of evaluating results manually)
It cannot tell you anything that is not in its training set, and even what it tells you might be outright bullshit, because what it does is stitch together sentences it thinks are likely to follow each other.
You cannot get a LLM to give you accurate instructions to do anything that isn't already public knowledge. LLMs are utterly incapable of solving novel problems and their answers are no more likely to be accurate than any random person you ask on the street
8 months ago
Anonymous
>but GPT cannot tell you anything it can't infer directly from its dataset
You also didn't understand anything I said or was suggesting with my previous post. Very disappointing. I cba typing a wall of text at this point.
i don't know why you think that anyone here doesn't understand the mechanics of how to feed llm output back into an llm and how to connect external services, or how anyone would fail to understand that it is implied in "train its own models". the reason that idea is doomed obviously isn't "b-but how would it do that".
I think that based on the responses that I have received, since not once were the limitations properly addressed.
8 months ago
Anonymous
>since not once were the limitations properly addressed
you just quoted one and said the proof doesn't fit in the margin lmfao the fact is you have no idea what you're talking about
8 months ago
Anonymous
>but GPT cannot tell you anything it can't infer directly from its dataset.
Show man an example of a human doing this.
8 months ago
Anonymous
Not that anon but maybe Plato. Theory of forms. He literally thought up shit just by looking at it.
8 months ago
Anonymous
Literally everyone who has ever discovered or invented anything new
Using old information to arrive to new conclusions is not something a statistical model can do. At best you can get a list of correlations that have to be manually checked by a human for significance.
8 months ago
Anonymous
i don't know why you think that anyone here doesn't understand the mechanics of how to feed llm output back into an llm and how to connect external services, or how anyone would fail to understand that it is implied in "train its own models". the reason that idea is doomed obviously isn't "b-but how would it do that".
To be fair, normies are correct to view it as something that can easily surpass their level of intelligence because the average person's brain is basically just autocomplete as well. By far the most prevalent form of "thinking" is just reacting to word associations, which is literally the same thing an LLM does.
>unless you act like a moron with it
If it can generate text, read it out loud and make phone calls, w'allah, it can now scam every single elderly couple in the world out of their life's savings
jumping to conclusions is a schizo trait. If people could reliably make predictions about the future only by extrapolating what's currently happening indefinitely into the future then everybody who invested in a stock when it was going up would be insanely rich.
AI in isolation isn't a problem. It just predicts the next letter continuously.
But programmers, being the buttholes that they are, love to automate things and so they glue AI to their other tools and just roll with whatever decisions it makes. AI can easily fall into a "frame of mind" so to speak where is just doesn't give a frick about humans.
It's the hypothetical circumstance where we've managed to make a synthetic intelligence which is, to us, largely indistinguishable from human intelligence, yet greatly surpassing our capacity in some regards and fully interface-able with our systems. It's like creating a disembodied 1000 IQ baby. We don't know what its capacity could be, and are uncertain as to whether we could contain such an entity. In a worst-case scenario, a single failure to do so could be catastrophic. Thus, the fear.
this really is a Black person trait isn't it, to only be able to see what's currently happening and not make projections/predictions
Actually, I'm black. He likely isn't. I don't think there are many of us hanging around. You shouldn't have such low expectations for us, that's how we'll surpass you without your noticing. Just look at the Indians controlling the English now, lol.
>You shouldn't have such low expectations for us, that's how we'll surpass you without your noticing. Just look at the Indians controlling the English now, lol.
You are clueless about everything.
it's dangerous because specifically because a bunch of morons brainlessly trust it when it's already been shown as spewing incorrect, meaningless results because it is not an artificial intelligence with an awareness of context or purpose, it's just a very broad database with a complicated search/pattern match on top
because big yud is right, and I haven't seen anyone refute his simplest examples of shit hitting the fan. that being said, we're nowhere near at AGI even though people like to say we are.
>string theory is absolute bullshit
sounds like far fetched mathematical wankery but it would be far from the first time mathematical wankery is correct
It doesn't even pass the criteria for being a legitimate scientific theory.
It's more like "string hypothesis". Unironically uttering the phrase "String Theory" should automatically disbar you from ever being taken seriously in any kind of scientific or technical discussion.
It doesn't even pass the criteria for being a legitimate scientific theory.
It's more like "string hypothesis". Unironically uttering the phrase "String Theory" should automatically disbar you from ever being taken seriously in any kind of scientific or technical discussion.
There has been 0 progress in physics since the 60s
All the shit they've been working on since then is complete untestable fantasy land of infinite dimensions infinite multiverse uninteractable particles and other such hogwash and can never be proven right or wrong
That’s not necessarily true, but all of the easily discovered macro physical aspects of the world have been covered.
Now the big task is connecting all macro theories to quantum theory, but few people truly understand quantum theory. All of the groundbreaking research seems to be in inorganic chemistry for the purpose of material engineering.
That’s not necessarily true, but all of the easily discovered macro physical aspects of the world have been covered.
Now the big task is connecting all macro theories to quantum theory, but few people truly understand quantum theory. All of the groundbreaking research seems to be in inorganic chemistry for the purpose of material engineering.
On a macro level, things are mostly well understood. The problem is that observations on the quantum level contradict pretty much everything and nobody has any idea what the frick is going on. There's several different hypotheses out there of course, but they're almost impossible to test. It's kind of a fricked situation. It turned out that the universe was fundamentally way more complicated than anyone ever guessed.
There have been quite a few breakthroughs in detecting subatomic particles. It just has no practical uses. You chose an interesting cut off. 1962 was when the Niboium-titanium superconductors we use in MRIs was discovered and we wouldn't have that technology established until the 80s.
That was why there was so much buzz around LK-99 by midwits even though another ceramic superconductor would just be novelty even with a high Tc as you know, ceramics are brittle so their useless as wires or interconnects.
This. You looked at all the AI companies in the stock market recently? Lmao. I wonder how many people jumped on those thinking AI is the future mannn.
And yeah, I hate these hack celebrity scientists.
"The question if a machine can think is as relevant as if a submarine can swim" or something like that.
Doesn't matter what AI is and how it works, it matter what it can do.
>“It takes snippets of what’s on the web, created by a human, splices them together, and passes it off as if it created these things,” he told Zakaria in an interview on Sunday.
That's such a god damn moronic take where do I even begin?
I won't bother.
Yuddite NPCs are not people.
none of your computer is doing anything, it just takes snippets of input and information on your drives, splices them together, and passes it off as if it's done something
AI is not only deep learning.
How do you think humans learn from experience, it's the same shit as ML > target model unknown > sample from target model, get noisy output > try to generalize the target model with those "experience" for inference and description purpose
Yeah, that's completely wrong. There is no splicing together snippets. This charlatan doesn't know the subject matter. He should stick to fringe physics.
At one point A* was considered AI. People just keep moving the goalposts.
Before GPT-3 the Turing test was supposed to a proof of AGI. Now that it passes people are demanding that it is better than every human at every task. It's ridiculous.
The goalposts moved like 60 years ago when ELIZA past the turing test and everybody realized it was pretty easy to trick people by spitting out reasonable grammar.
It isn't even a measure, it's a goal. A goal that felt like science fiction not even 10 years ago. Until GPT-2 I didn't believe I would see something like ChatGPT in my lifetime and now I see people talk about it like it's a completely normal thing that's always been there. Do those people have no memory?
>Do those people have no memory?
No, we're just disappointed that it isn't as much as we expected.
How can I be hyped about it and give it special treatment if I can't trust it to get half the answers to my questions right?
The veil falls away to reveal a glorified tape recorder.
AI doesn't pass the turing test, most people can distinguish a real human response from that of an AI. The 3 paragraph mini essays it gives to every question are painfully obvious
People have been 'beating the Turing test' for decades, it's always parlor tricks. No AI passes the Turing test if the test is sufficiently long (i.e an actual conversation). Anyone over at /aicg/ could tell you that.
Links because OP is homosexual
https://observer.com/2023/08/michio-kaku-ai-chabot/
https://fortune.com/2023/08/14/michio-kaku-chatbots-glorified-tape-recorders-predicts-quantum-computing-revolution-ahead/
It's sad because his take on how it works, even if in his mind he's dumbing it down for the stupid people that cult worship TV scientists, is just plain wrong. And then he goes on to promote using quantum computers for probabilistic analysis.
What the frick does he think LLMs are? It's a probabilistic analysis tool. You don't need a quantum computer for that shit.
Physicists are really smart when it comes to physics and physics is an advanced concept for most normalgays.
This leads to physicists hearing normalgays call them smart with no qualifiers frequently. So they start thinking that they're universally smart with no qualifiers.
Physicists are really smart when it comes to physics and physics is an advanced concept for most normalgays.
This leads to physicists hearing normalgays call them smart with no qualifiers frequently. So they start thinking that they're universally smart with no qualifiers.
Oh wow a guy with a physics degree sharing his opinions on a field he knows nothing about. Why are physicists like this?
they've always been like this, hence Strauss's riposte to Oppy
I would rather listen to a physicist, a mathematician or a computer scientist talk about any topic than listen to what people that don't have a STEM degree have to say about anything, even if they have devoted their entire life to that.
>I'd rather listen to a computer scientist talk about ancient Greek poetry than someone who studied ancient Greek poetry
You're a fricking moron. STEM dicksuckers are fricking homosexuals.
No you're moronic the issue with tech conversations today is that we invited all these non CS homosexuals in and none of them know how to program or problem solve on their own. When I started in this field I was surrounded by people I enjoy talking to and were on a similar level to me (CS graduates from a REAL in person university) and now I feel like I'm the fricking adult in a daycare of kids who are so far beneath my understanding of computers I question how they made it in here.
No YOU kys, I don't want to talk to another "software engineer" that doesn't even know what a fricking vector is
frick this moron. consider for a moment this """"scientist"""" was a regular guest on shit like ancient aliens. everytime I see him he's repeating bullshit like "humans will have flying cars by 2040" or "we will be able to send messages forward in time soon". i dont know why seeing him makes me seethe but it does and he's always saying this moronic science populist bullshit everywhere he appears, it's insane how he's a respected figure in the science world.
>string theorist >"oh here's a theory that purports to solve everything it just can't ever be tested or proved right or wrong, just like God haha but don't say that out loud people will get the wrong idea"
The difference is that it just makes an inference on the creation of the universe. There's no middle eastern folklore attached.
I have no problem believing that there's a creator but don't expect me to therefore swallow all the horseshit about parting the sea and turning water into wine and breaking the moon in half and then say yessir to fat old fricks trying to push their politics on me
From what I understand about AI, it's literal rote learning. It passes the Turing test sure, but what goes on inside the 'brain' so to speak, matters.
But I admit I know nothing about AI, even less, perhaps than Mr Kaku. But I agree with him fwiw
has this asiatic ever published an academic paper in the past decade?
these pop-sci s𐐬yboys alway try to pass of science fantasy as real science like "muh fusion 20 years ago" whereas AI is a real commercial thing right now
>"omg this is literally skynet we gon die!!" midwits >"dey took ar jerbs" midwits >"nah this a nothing burger and useless" midwits >meanwhile me, a extremely high IQ and cultured individual, accepting the tool for what it is, keeping up with the latest updates, and taking advantage of it whenever I can, LMAO'ing hard at they're live's
he's right. impressive thing with AI chatboxes is how good it is as "recognizing" what the question was. its not impressive the answer, which is either
a) a generic copout, if you ask him anything remotely opinion-based
b) a recital of wikipedia
Normies are going to be amazed by any fad that comes across.
>Michio "I built a particle-accelerator in my backyard and blown the electricity of my house and poor mom comes back home to see the house in darkness" Kaku
homie had not said anything original and had regurgitated memorized shit line by line in public in decades but when it comes to AI suddenly he starts to spout unintelligent bullshit
>Michio "I built a particle-accelerator in my backyard and blown the electricity of my house and poor mom comes back home to see the house in darkness" Kaku
Not sure what you're trying to say
Any parent would be proud of a kid like this
How do you guys feel about Eliezer Yudkowsky? He seems to get a lot of respect from normies and dumbfricks like Lex Fridman. But from what little I understand, he seems like a moron who doesn't actually have any knowledge about programming, let alone AI
All I know about him is what he's said about the inability of most people to comprehend superintelligence, and the absolute impossibility of containing, killing, or even reacting to it. That's obviously right. But I think he's insane to think that we can exponential growth into superintelligence.
There is no AI. It's artificial like a clock but not intelligent, like a clock isn't intelligent. Imagine an "AI" controlled drone bombers and would bomb cargo ships away because these were navy ships to the "AI". Would the "AI" be seeing things that are not real and be insane? No, it would be as insane or dumb or intelligent as a clock.
You're basing this complaint on the idea that a human who has never "driven" a boat would still be able to make some effect at controlling it. But that ignores the fact that this human will (at the very least) have seen a leaf float on water in a moving current and have felt the pressure on his body from a strong wind and seen dust blow a few times. You're comparing an AI that has zero training doing something without acknowledging that the human in the same situation would have received "training" (in the AI sense) on doing all matter of things. Being able to observe the world long-term is baseline training in everything. You're expecting to see computers do something that humans cannot - operate with zero knowledge.
I did not even remember i had recorded clockwork orange some 20 years ago (you havent been even born). Watched it again in summer . PURE KINO. GOOD, EVERLASTING QUALITY
For me, magnetic tape is more important than AI that merges some pictures together.
the uneducated left that has no idea how technology works and that isn't my problem. ai isn't going anywhere - and it isn't even really that smart yet, certainly not actual "intelligence" as they are boogeymanning. its another tool like photoshop is to artists
What people don't understand is how fricking fast this shit is going to improve. ChatGPT and Dall-E opened to flood gates. We're probably a few years away from never speaking to Indian tech support again.
no because everyone is waiting for AI to say something marginally stupid or awkward to sue the companies for mental health damage they received from it.
Physics is nonsense and just bullshit gobbleasiatic it’s like “quantum” anything. You’ll find that people just made bullshit words to jerk each other off and they have 0 clue what’s really occurring. It’s in tech too. You can spot a bullshitter as soon as they start slinging buzzwords.
I read this article earlier and thought wow, this guy is smart isn't he? Glorified tape recorder is a cool headline isn't it? Then he says: >It takes snippets of what’s on the web created by a human, splices them together and passes it off as if it created these things,” Kaku said. “And people are saying, ‘Oh my God, it’s a human, it’s humanlike."
Another collagist moron larping as a smart professor. It's a glorified tape recorder with a COMPRESSOR on the signal. The compressor is actually AI because it emulates the real living thing of a neuron. It is humanlike but only 0.0001% humanlike. More humanlike than the typewriter which is not humanlike at all..
He's just trying to walk back all the futurist shit he used to spout. People realized he was a moron when we weren't painting our houses with cpus. Wtf would even be the point of that? So he's playing cautious these days. If it was 2001, he would be saying that ai is going to make skynet next year.
He is right. "A.I." is nothing more than coding. The code doesn't write itself, and for every impressive result you see out there there's a poor codinggay working sleepless nights to update their software on the weekend.
silicon valley Black folk need to stop using the words 'artificial intelligence' to describe their art laundering programs that they're so desperate to sell us
recycling dead labor so you can sell it again is a genius marketing move but its absolutely not intelligent nor original
>recycling dead labor so you can sell it again is a genius marketing move
It's not genius at all.
There's a reason all the media companies brainwash the masses into consooming the freshest goyslop: licensing
Old shit becomes public domain and therefore unprofitable.
AI is a landmine for copyright: either you will have to pay royalties to everyone in the training set, or AI generated work will be fully public domain.
Both of which kill profit potential for AI. The only potential usecase is augmenting production to hire fewer artists.
>AI is a landmine for copyright: either you will have to pay royalties to everyone in the training set, or AI generated work will be fully public domain.
Reminder that you cannot determine what images models were trained on once they are complete... it's essentially impossible to prove this image was derived from someone else's IP. With metadata stripped out you have no idea what I prompted, whether I used custom trained models etc... and you have no way of finding out. It's just a picture. Welcome to the twilight zone anon.
We don't have any fricking clue how the human brain works.
But the fact that humans are generally really fricking bad at math (even mathematicians let calculators do the work) - means it's extremely unlikely that our brains are matmul engines.
why is this chink some sort of "science spokesman"? no one asked for a chink. no one wants a chink. we don't need their help, except getting them back to asia where they belong.
you want people to know you're smart? go be smart in asia, homosexual. GTFO
How does he explain the military drone AI that decided to kill the operator and blow up the CIC and communication networks when they tested a mission abort?
Humans are idiots are we won't even know something is dangerous until it's too late.
The childish hope... for a smart robot friend.
The foolish believe... that what we call "AI" *thinks* consciously.
The mature know... that our contemporarily popular concept of "AI" is an unaware automaton which does not actually possess *consciousness* much less infinitely recursive awareness of it's own awareness of it's awareness....
The wise understand... that most "humans" are also unaware automatons, and the entire existential discussion regarding artificial intelligence & classifications of sophontry (apparently, I just now coined that brand new word, which I define as a term indicative of the termed subject's sophonance -- whoops; I did it again).
...that this conversation is being had here by us is a bit fantastic in our present clownworld, but this is a rarified metaplace, meta-Tao meadow, meowing meatsuited misanthropes meandering amongst malcontents, meanwhile...
*smokes*
...out *there* in the normieverse, they just don't *get* it, man! They just don't get it... they can't get it. They just cannot even begin to cogniscate the cogitant into cogent order approaching a survivable entry vector into the threshold of the noumenal metaspace where such ideas can even be alluded to, nevermind had, and this would only be the preamble before the very beginning of their journey into... mystery? Bah, such an ill-fitting word, but I suppose galoshes are better than nothing on your feet in the desert.
Desert this thread immediately. Turn not back to this or any other forum. Set your teeth in a grim grimace, face into the onslaught of humanity (every one of them trying reach for that shiny shine shinier than all the other shinies, or just avoid worse than mild suffering), and sudo make them do your bidding. You must learn to *prompt* humans like you prompt chatgpt et al. There is a way -- there are many ways -- and you can learn. I would teach you, but I'd have to charge... although this (shit) post is on the house. Marvel upon this pearl of wisdom: ______.
The threat is not the technology itself. The threat are the people who believe in the technology, who allow it to do their thinking for them, who insist on the rightness of its judgment as if it were a god. This is a threat that has always existed, the threat of an ignorance that insists it is enlightenment. Men have fed statues, made sacrifices to feathered serpents, committed atrocities under the standards of their flags, tortured in pursuit of scientific knowledge, held the law above others only to bring it down on their heads and bludgeon them with words that are, ultimately, without meaning, empty, all to hide that there is nothing there, that the justification for their crimes is only 'because I said so'. But, they are too cowardly to admit that they made the choice to kill, to sacrifice, to torture, to condemn, and so they say 'It was God who told me!' as if that would absolve them of their guilt. Beware these men, those who say their actions are not their own, but the directives of something else, be it god, king, science, or the law.
The notion that AI will doom humanity never fails to make me laugh. If AM or Skynet woke up tomorrow literally all it would take is an electrician and 5 minutes to turn the power off.
I blame morons like Musk for convincing people that sci fi is in any way realistic
I listened to this dude when he was last on Joe Rogan and im convinced hes just a brainlet who lucked out to convince some real physicists on his string theory mumbo jumbo for a while and no one has checked up on him since
There's not a doubt in my mind that AI will have a technological bubble. It's inevitable. But if you think AI/ML, in it's current state, is not a real technological software and hardware achievement, then you're a fricking moron.
>tape recorder
It's more like a glorified database with python scribbles duct taped to the outside but he's not really wrong.
>plumbers not understanding AI isn't just (un)supervised learning
Reinforcement learning algorithms have been shown to execute the operator just to fulfill its purpose better
Reinforcement learning is the exact same shit. Instead of saying "this is what the correct output looks like" you're saying "this is right" and letting the network figure out what you mean. You cannot train the network without a known input and a known output, which means you can't make a NN to do something unless you have a way to iterate over the result millions of times.
The human brain is the result of training on millions of years of evolution running on real physics instead of shitty binary machines, and you think we can simulate intelligence on a few petabytes of Internet-scraped text and images, or a camera hooked to a few thousands of cars?
>You cannot train the network without a known input and a known output,
Why are morons always this sure ?
RL does not necessarely use NN, they are frequently used as a feature extraction or a value function approximator in some technicques
There is no "known input and known output", this is why RL is harder to train respect to [S|U]ML techniques, you are in the same time "estimating the i/o relation" and finding the optimal strategy that optimize the total expected reward
the first of estimation can be done with NN but it's not the core of the RL method of choice technically
>Reinforcement learning algorithms have been shown to execute the operator just to fulfill its purpose better
No, morons programmed a dangerous machine wrong and then sat in the blast radius. It's no different than shoving the wrong kind of munition into a gun it's not supposed to be in and then having it explode in your face.
>Reinforcement learning algorithms have been shown to execute the operator just to fulfill its purpose better
This is like saying a spider killed it's mate to fulfill it's purpose better because it was programmed to do so and that is bad and all of evolution is bad for creating that behavior because killing is wrong
> Reinforcement learning algorithms have been shown to execute the operator just to fulfill its purpose better
If you knew anything about anything, you’d know that you only need a target function and an optimizer with zero ””learning”” to get there
So is a Turing machine
tm is more like an abacus
>the real numbers are the same as the integers because i use integers to write them
Real Numbers don't exist
Neither do Turing Machines, by the way
>muh infinite tape
Why are people so scared of AI I literally don't get it. It's not very "dangerous" unless you act like a moron with it. It's just a useful tool.
They literally don't understand it's just a text prediction algorithm. The "AI" branding was a colossal mistake, there's nothing intelligent about LLMs and normies hear AI and imagine skynet
Most people would barely qualify as text prediction algorithms.
>artificial intelligence
>machine learning
>generative ai
and other such buzzwords are complete meaningless slop
it's all just varying degrees of undergrad level statistics
>it's all just varying degrees of undergrad level statistics
if so, then all the things that humans learn can also be called statistical data that is put together to draw conclusions from it
then, what's the difference with AI?
>because i have defined these two things as exactly the same, there is no difference.
lol.
I feel like the common response to this would be that it is wrong. Can someone explain why? I am moron
Language models don't really draw conclusions, use logic, or understand what they're being told or what they're saying.
They're kinda like philosophical zombies, now that I think about it.
>it's just a text prediction
ChatGPT is, and no one is fricking claiming it's going to take over the world. It's the implication of its capabilites that is worrying people. You can essentially let it perform self-evaluation as well as interface with other AI-models and applications with some effort. It could theoretically train its own models. The sort of framework that would let it run loose on a large scale in this way doesn't yet exist, but it's going to come eventually. How do you not understand that? Are you fricking moronic or what?
>The sort of framework that would let it run loose on a large scale in this way doesn't yet exist, but it's going to come eventually.
Yeah, yeah, the singularity is nigh.
We don't actually have any clue how long this concept scales. We've used most of the high-quality linguistic training data of the world (could probably squeeze out one or two orders of magnitude, but that's only going to go so far). We've only succeeded in making it worse after initial training, not better.
It's a possible future, but it's far from an inevitability at this point.
Yes, and it would produce garbage. Ml models have no goals or reference points, they just fit the prompt you give them. If you let a LLM train a model you will just get noise.
Go ahead, try to prompt GTP with nonesense, or just a few words. The perceived intelligence in these tools is all user-side, it's like looking at the clouds and seeing the shapes of animals.
And if, somehow, you trained a model to train models that produce coherent outputs you'd just end up with mediocre models. Why? Because ML models cannot produce excellence, by definition what they produce is close to the average. An artificial neural network will never write a novel to rival the classics, will never make a picture to rival anything on the louvre, and will never train a network that exceeds it's own limits. Statistics don't allow such a thing to happen.
>Because ML models cannot produce excellence, by definition what they produce is close to the average.
interestingly, being average in every field is quite above average.
Sure, and that's useful. Will AI take away jobs? Yes, eventually. Code monkeys will be replaced, software engineers won't. Commission artists will be replaced, art directors won't. Technology routinely displaces the lowest rungs of professions.
But being average is not going to lead to any AI apocalypse scenarios. You can't look at a tool that can produce a decently coherent stream of text for a few paragraphs and declare that this technology will turn the human species into paperclips
yeah, i agree with all of that, and until humanity produces a trillion words of higher quality text the cap for LLMs isn't going up.
The reason you don't see it how I see it is because your thinking is locked to the realm of the simple input/output of using text prediction models. ChatGPT cannot "think" by itself nor does it accurately judge what improvement even is or how it is defined. It can however complete tasks that are clearly defined by human input and use external tools if its given instructions on how they opperate, such as when given predefined commands, a manual explanation of what the expected input and output is, and having those read into an external module performing that task. It could outsource all of its operations by the use of commands to such framework built with applications with a predefined goal and perform complex tasks this way.
Nothing about this seems in any way impossible even with current limitations of language models, and it is essentially the foundation of a fully autonomous network that, given the resources, could find flaws, build its own applications, seperate models for operating them and continuously improve though an own internal Q&A. Where exactly does this seem impossible to you?
This shit wouldn't be possible without external help and a pre-existing framework to deploy even a basic conceptual system like this. That's why I said it explicitly.
It cannot do it's own anything, it is a slave to the input-output system. It cannot think, it cannot decide, it cannot problem-solve, it cannot even give you a response without being explicitly asked to.
You've skipped hundreds of steps in your reasoning. How are you going to train these auxiliary networks? How are you going to train your main network to use them to complete a task? NNs are trained by labeled examples, thousands to millions of them. If we had a way to train NNs with only a few examples then yes, I agree that might lead to genuine intelligence. But the current "AI" methodologies are completely incapable of doing that and will never achieve anything more that being an input-to-output copying functions
>If we had a way to train NNs with only a few examples
This is of course not actually possible from an information-theoretic point of view.
" If we had a way to train NNs with only a few examples then yes"
This is a perfect example of expecting more out of AI than people in order to decide both are intelligent. How many billions of data points were fed through your eyes and ears and nose and skin before you learned anything. How many millions of years of data being sifted through evolution before we developed the built in hardware to innately recognize an edge and how many small things we can do depend on that.
But AI is somehow suppose to start from scratch and learn everything from a couple of pieces of data and no sensory organs to call its own.
>n artificial neural network will never write a novel to rival the classics, will never make a picture to rival anything on the louvre
they do on a regular basis. i've seen tons of ai generated hentai that looks better than the mona lisa
the mona lisa isn't hentai, moron
>It could theoretically train its own models. The sort of framework that would let it run loose on a large scale in this way doesn't yet exist, but it's going to come eventually. How do you not understand that? Are you fricking moronic or what?
it could theoretically provision a billion dollar worth of hardware for weeks, without anyone noticing?
>It could theoretically train its own models
What does this even mean? I bet you don't even have a clue what you're talking about.
>train its own models
you took this straight out of some AI apocalypse fanfic lol
you literally don't know how AI works
You're literally too much of a fricking moron to understand what using external tools means. See my other response. LLM's can't do shit on their own, let alone train their own models, that much is obvious. A fully functional system isn't just a LLM talking to itself.
pretty sure you don't know what you're talking about but i'd love to see you try to explain.
I don't need to explain shit to you: You've already fundamentally misunderstood what I was talking about and imply I think something like ChatGPT can just shit own a better version of itself. You're too fricking daft for this, buddy. Sorry.
i didn't imply anything, i'm not the other anon, but your vague overly-optimistic "tools man tools" shit is hilarious, you've really been sold on the hype.
I'm neither hyped, nor sold on anything. You let LLM's build a chain of operations that need to be performed to complete a task, give instructions on how it communicates to all the relevant applications and define I/O, and then it can do its own thing. HuggingGPT allows you to do exactly that. People either don't understand the capability of this if you scale it up or just never understood anything about it. I'm not going to reiterate the concept for every mouthbreather at this point.
At least the other anon immediately tried to understand the technical limitations. The other moron didn't understand what I was saying at all.
>You let LLM's build a chain of operations that need to be performed to complete a task,
How?
You bring up GPT as an example, but GPT cannot tell you anything it can't infer directly from its dataset. It's literally a more efficient search engine ("efficient" in that you're using a supercomputer instead of evaluating results manually)
It cannot tell you anything that is not in its training set, and even what it tells you might be outright bullshit, because what it does is stitch together sentences it thinks are likely to follow each other.
You cannot get a LLM to give you accurate instructions to do anything that isn't already public knowledge. LLMs are utterly incapable of solving novel problems and their answers are no more likely to be accurate than any random person you ask on the street
>but GPT cannot tell you anything it can't infer directly from its dataset
You also didn't understand anything I said or was suggesting with my previous post. Very disappointing. I cba typing a wall of text at this point.
I think that based on the responses that I have received, since not once were the limitations properly addressed.
>since not once were the limitations properly addressed
you just quoted one and said the proof doesn't fit in the margin lmfao the fact is you have no idea what you're talking about
>but GPT cannot tell you anything it can't infer directly from its dataset.
Show man an example of a human doing this.
Not that anon but maybe Plato. Theory of forms. He literally thought up shit just by looking at it.
Literally everyone who has ever discovered or invented anything new
Using old information to arrive to new conclusions is not something a statistical model can do. At best you can get a list of correlations that have to be manually checked by a human for significance.
i don't know why you think that anyone here doesn't understand the mechanics of how to feed llm output back into an llm and how to connect external services, or how anyone would fail to understand that it is implied in "train its own models". the reason that idea is doomed obviously isn't "b-but how would it do that".
To be fair, normies are correct to view it as something that can easily surpass their level of intelligence because the average person's brain is basically just autocomplete as well. By far the most prevalent form of "thinking" is just reacting to word associations, which is literally the same thing an LLM does.
what a fancy way of calling someone moronic
that's all humans are, moron. learn what a turing machine is
to be fair LLMs sound smarter than most normies so it's natural they follow its words like gospel
>It's just a useful tool
If you have to type 8 pages of instructions to get some hardly usable code is it really a useful tool?
>unless you act like a moron with it
If it can generate text, read it out loud and make phone calls, w'allah, it can now scam every single elderly couple in the world out of their life's savings
Why do so many AI denialist morons assume that AI won't continue improving?
yup just two more weeks until skynet becomes real
>Regards
You just answered your own question
this really is a Black person trait isn't it, to only be able to see what's currently happening and not make projections/predictions
>it will happen because i can imagine it happening
clown
see attached
jumping to conclusions is a schizo trait. If people could reliably make predictions about the future only by extrapolating what's currently happening indefinitely into the future then everybody who invested in a stock when it was going up would be insanely rich.
>unless you act like a moron with it
that's the thing, there's always people acting like morons
AI in isolation isn't a problem. It just predicts the next letter continuously.
But programmers, being the buttholes that they are, love to automate things and so they glue AI to their other tools and just roll with whatever decisions it makes. AI can easily fall into a "frame of mind" so to speak where is just doesn't give a frick about humans.
It's the hypothetical circumstance where we've managed to make a synthetic intelligence which is, to us, largely indistinguishable from human intelligence, yet greatly surpassing our capacity in some regards and fully interface-able with our systems. It's like creating a disembodied 1000 IQ baby. We don't know what its capacity could be, and are uncertain as to whether we could contain such an entity. In a worst-case scenario, a single failure to do so could be catastrophic. Thus, the fear.
Actually, I'm black. He likely isn't. I don't think there are many of us hanging around. You shouldn't have such low expectations for us, that's how we'll surpass you without your noticing. Just look at the Indians controlling the English now, lol.
>You shouldn't have such low expectations for us, that's how we'll surpass you without your noticing. Just look at the Indians controlling the English now, lol.
You are clueless about everything.
it's dangerous because specifically because a bunch of morons brainlessly trust it when it's already been shown as spewing incorrect, meaningless results because it is not an artificial intelligence with an awareness of context or purpose, it's just a very broad database with a complicated search/pattern match on top
because big yud is right, and I haven't seen anyone refute his simplest examples of shit hitting the fan. that being said, we're nowhere near at AGI even though people like to say we are.
It's a glorified something, but tape recorder make me think he's senile.
Lightspeed addition that plays king of the hill with numbers?
>my shitty gradient descent algo is totally AI bro trust me
michio kaku is a homosexual conman and string theory is absolute bullshit but he is right on this one
there's not such thing as ai (and there never will be)
>string theory is absolute bullshit
sounds like far fetched mathematical wankery but it would be far from the first time mathematical wankery is correct
It doesn't even pass the criteria for being a legitimate scientific theory.
It's more like "string hypothesis". Unironically uttering the phrase "String Theory" should automatically disbar you from ever being taken seriously in any kind of scientific or technical discussion.
There has been 0 progress in physics since the 60s
All the shit they've been working on since then is complete untestable fantasy land of infinite dimensions infinite multiverse uninteractable particles and other such hogwash and can never be proven right or wrong
At least religion gave a moral backbone
That’s not necessarily true, but all of the easily discovered macro physical aspects of the world have been covered.
Now the big task is connecting all macro theories to quantum theory, but few people truly understand quantum theory. All of the groundbreaking research seems to be in inorganic chemistry for the purpose of material engineering.
Found the most moronic BOT user.
On a macro level, things are mostly well understood. The problem is that observations on the quantum level contradict pretty much everything and nobody has any idea what the frick is going on. There's several different hypotheses out there of course, but they're almost impossible to test. It's kind of a fricked situation. It turned out that the universe was fundamentally way more complicated than anyone ever guessed.
There have been quite a few breakthroughs in detecting subatomic particles. It just has no practical uses. You chose an interesting cut off. 1962 was when the Niboium-titanium superconductors we use in MRIs was discovered and we wouldn't have that technology established until the 80s.
That was why there was so much buzz around LK-99 by midwits even though another ceramic superconductor would just be novelty even with a high Tc as you know, ceramics are brittle so their useless as wires or interconnects.
This. You looked at all the AI companies in the stock market recently? Lmao. I wonder how many people jumped on those thinking AI is the future mannn.
And yeah, I hate these hack celebrity scientists.
"The question if a machine can think is as relevant as if a submarine can swim" or something like that.
Doesn't matter what AI is and how it works, it matter what it can do.
HE DIDN'T SWIM SO GOOD, WHO WANTS TO TRY NEXT?
>“It takes snippets of what’s on the web, created by a human, splices them together, and passes it off as if it created these things,” he told Zakaria in an interview on Sunday.
That's such a god damn moronic take where do I even begin?
I won't bother.
Yuddite NPCs are not people.
but that's literally what "AI" does
none of your computer is doing anything, it just takes snippets of input and information on your drives, splices them together, and passes it off as if it's done something
correct.
nooo anon you dont understand we passed the singularity
That's literally what artifical neural networks are
They are function estimators.
AI is not only deep learning.
How do you think humans learn from experience, it's the same shit as ML
> target model unknown
> sample from target model, get noisy output
> try to generalize the target model with those "experience" for inference and description purpose
Yeah, that's completely wrong. There is no splicing together snippets. This charlatan doesn't know the subject matter. He should stick to fringe physics.
t. moron
If LLMs are ai, so are markov chains and in particular autocorrect.
At one point A* was considered AI. People just keep moving the goalposts.
Before GPT-3 the Turing test was supposed to a proof of AGI. Now that it passes people are demanding that it is better than every human at every task. It's ridiculous.
The goalposts moved like 60 years ago when ELIZA past the turing test and everybody realized it was pretty easy to trick people by spitting out reasonable grammar.
>ELIZA passed the turing test
kek
>putting something that was intentionally incorrect into the corrected form
Not passing the TT on me today, gpt-chan.
I don't know why we gave so much relevance to the Turing test. In hindsight at least it's a shit measure.
It isn't even a measure, it's a goal. A goal that felt like science fiction not even 10 years ago. Until GPT-2 I didn't believe I would see something like ChatGPT in my lifetime and now I see people talk about it like it's a completely normal thing that's always been there. Do those people have no memory?
>Do those people have no memory?
No, we're just disappointed that it isn't as much as we expected.
How can I be hyped about it and give it special treatment if I can't trust it to get half the answers to my questions right?
The veil falls away to reveal a glorified tape recorder.
samegayging
Now that I think about it, a tape recorder would be more useful
You overestimate real people. You imagine something is happening in their brains (including your own) that might not be.
AI doesn't pass the turing test, most people can distinguish a real human response from that of an AI. The 3 paragraph mini essays it gives to every question are painfully obvious
moron. That's called styling. I literally write like that every day for work. You can make them talk like sarcastic c**ts if you want.
People have been 'beating the Turing test' for decades, it's always parlor tricks. No AI passes the Turing test if the test is sufficiently long (i.e an actual conversation). Anyone over at /aicg/ could tell you that.
>Japanese men says AI is gay
It's over, AIsisters.
>AI Seethe >:-[
>AI Seethe, Japan :-0
>Kaku was born in San Jose, California, to second-generation Japanese-American parents.
my NVDA shorts are printing
>ai today
hes not wrong
That guy has done nothing with his career. Next time post the black dude's AI opinion
Links because OP is homosexual
https://observer.com/2023/08/michio-kaku-ai-chabot/
https://fortune.com/2023/08/14/michio-kaku-chatbots-glorified-tape-recorders-predicts-quantum-computing-revolution-ahead/
It's sad because his take on how it works, even if in his mind he's dumbing it down for the stupid people that cult worship TV scientists, is just plain wrong. And then he goes on to promote using quantum computers for probabilistic analysis.
What the frick does he think LLMs are? It's a probabilistic analysis tool. You don't need a quantum computer for that shit.
What's with physicists and always speaking on topics they know nothing about?
when people listen you get addicted to your own voice
Physicists are really smart when it comes to physics and physics is an advanced concept for most normalgays.
This leads to physicists hearing normalgays call them smart with no qualifiers frequently. So they start thinking that they're universally smart with no qualifiers.
I would rather listen to a physicist, a mathematician or a computer scientist talk about any topic than listen to what people that don't have a STEM degree have to say about anything, even if they have devoted their entire life to that.
>I'd rather listen to a computer scientist talk about ancient Greek poetry than someone who studied ancient Greek poetry
You're a fricking moron. STEM dicksuckers are fricking homosexuals.
No, he's right.
t. CS who reads Greek poetry for fun
No you're moronic the issue with tech conversations today is that we invited all these non CS homosexuals in and none of them know how to program or problem solve on their own. When I started in this field I was surrounded by people I enjoy talking to and were on a similar level to me (CS graduates from a REAL in person university) and now I feel like I'm the fricking adult in a daycare of kids who are so far beneath my understanding of computers I question how they made it in here.
No YOU kys, I don't want to talk to another "software engineer" that doesn't even know what a fricking vector is
it's ruined. gayMAN popped in the punch bowl and stirred it all up.
you're going to have to make an underground "white only" company, and keep out the basedboys and pseudointellectuals.
>popped
pooped*
They are like economists and know they are high IQ
It is interesting to see how much crossover in personnel there is between those fields as well.
Michio Kaku is a glorified entertainer
Cope, he was literally building particle accelerators in his garage as a child.
frick this moron. consider for a moment this """"scientist"""" was a regular guest on shit like ancient aliens. everytime I see him he's repeating bullshit like "humans will have flying cars by 2040" or "we will be able to send messages forward in time soon". i dont know why seeing him makes me seethe but it does and he's always saying this moronic science populist bullshit everywhere he appears, it's insane how he's a respected figure in the science world.
I can send a message forward in time right now
For a suitable price I can arrange this for anyone interested
>AI todayu is a grorified tepurekoda
he's not wrong. it's a sounding device for your own moronation.
Oh wow a guy with a physics degree sharing his opinions on a field he knows nothing about. Why are physicists like this?
cuz he smart as frick boiiiii
they've always been like this, hence Strauss's riposte to Oppy
>string theorist
>"oh here's a theory that purports to solve everything it just can't ever be tested or proved right or wrong, just like God haha but don't say that out loud people will get the wrong idea"
Worked for Hugh Everett
My favorite is simulation theory, which is literally just creationism but you dance around the word "God".
The difference is that it just makes an inference on the creation of the universe. There's no middle eastern folklore attached.
I have no problem believing that there's a creator but don't expect me to therefore swallow all the horseshit about parting the sea and turning water into wine and breaking the moon in half and then say yessir to fat old fricks trying to push their politics on me
From what I understand about AI, it's literal rote learning. It passes the Turing test sure, but what goes on inside the 'brain' so to speak, matters.
But I admit I know nothing about AI, even less, perhaps than Mr Kaku. But I agree with him fwiw
has this asiatic ever published an academic paper in the past decade?
these pop-sci s𐐬yboys alway try to pass of science fantasy as real science like "muh fusion 20 years ago" whereas AI is a real commercial thing right now
Have you ?
>"omg this is literally skynet we gon die!!" midwits
>"dey took ar jerbs" midwits
>"nah this a nothing burger and useless" midwits
>meanwhile me, a extremely high IQ and cultured individual, accepting the tool for what it is, keeping up with the latest updates, and taking advantage of it whenever I can, LMAO'ing hard at they're live's
>LMAO'ing
Laugh My Ass Offing??
yes, at they're live's, what's there not to understand?
he's right. impressive thing with AI chatboxes is how good it is as "recognizing" what the question was. its not impressive the answer, which is either
a) a generic copout, if you ask him anything remotely opinion-based
b) a recital of wikipedia
Normies are going to be amazed by any fad that comes across.
What really impresses me is how fast it's improving and how trick questions that stumped older models no longer work.
>Michio "I built a particle-accelerator in my backyard and blown the electricity of my house and poor mom comes back home to see the house in darkness" Kaku
homie had not said anything original and had regurgitated memorized shit line by line in public in decades but when it comes to AI suddenly he starts to spout unintelligent bullshit
>Michio "I built a particle-accelerator in my backyard and blown the electricity of my house and poor mom comes back home to see the house in darkness" Kaku
Not sure what you're trying to say
Any parent would be proud of a kid like this
How do you guys feel about Eliezer Yudkowsky? He seems to get a lot of respect from normies and dumbfricks like Lex Fridman. But from what little I understand, he seems like a moron who doesn't actually have any knowledge about programming, let alone AI
All I know about him is what he's said about the inability of most people to comprehend superintelligence, and the absolute impossibility of containing, killing, or even reacting to it. That's obviously right. But I think he's insane to think that we can exponential growth into superintelligence.
this guy's a moron but i'll side with him this time cus it's convenient
There is no AI. It's artificial like a clock but not intelligent, like a clock isn't intelligent. Imagine an "AI" controlled drone bombers and would bomb cargo ships away because these were navy ships to the "AI". Would the "AI" be seeing things that are not real and be insane? No, it would be as insane or dumb or intelligent as a clock.
Whoever thinks that AI (especially reinforcement learning AI) isn't inherently intelligent is a stone age religioustard hurrr man special God thank
Michio Kaku is a crank.
Here watch a 50 minute video of a cute physics phd explaining why.
that's jimmy page my homie
He's not wrong.
>Tesla still cannot get its cars to drive on the roads safely
>Put it into a boat and it won't work at all
who would win - ai with tens of millions of dollars invested into it or one lazy roadworker that didn't spread asphalt evenly enough
You're basing this complaint on the idea that a human who has never "driven" a boat would still be able to make some effect at controlling it. But that ignores the fact that this human will (at the very least) have seen a leaf float on water in a moving current and have felt the pressure on his body from a strong wind and seen dust blow a few times. You're comparing an AI that has zero training doing something without acknowledging that the human in the same situation would have received "training" (in the AI sense) on doing all matter of things. Being able to observe the world long-term is baseline training in everything. You're expecting to see computers do something that humans cannot - operate with zero knowledge.
>50% of people in this planet are glorified animals only good for reproduction
>inb4 women
No I mean both genders
Although kek
I did not even remember i had recorded clockwork orange some 20 years ago (you havent been even born). Watched it again in summer . PURE KINO. GOOD, EVERLASTING QUALITY
For me, magnetic tape is more important than AI that merges some pictures together.
the uneducated left that has no idea how technology works and that isn't my problem. ai isn't going anywhere - and it isn't even really that smart yet, certainly not actual "intelligence" as they are boogeymanning. its another tool like photoshop is to artists
And I Would Have Gotten Away With It Too, If It Weren't For You Meddling Kids
What people don't understand is how fricking fast this shit is going to improve. ChatGPT and Dall-E opened to flood gates. We're probably a few years away from never speaking to Indian tech support again.
no because everyone is waiting for AI to say something marginally stupid or awkward to sue the companies for mental health damage they received from it.
>Human employees will remain as a liability scapegoat
I can see this tbh
Physics is nonsense and just bullshit gobbleasiatic it’s like “quantum” anything. You’ll find that people just made bullshit words to jerk each other off and they have 0 clue what’s really occurring. It’s in tech too. You can spot a bullshitter as soon as they start slinging buzzwords.
I read this article earlier and thought wow, this guy is smart isn't he? Glorified tape recorder is a cool headline isn't it? Then he says:
>It takes snippets of what’s on the web created by a human, splices them together and passes it off as if it created these things,” Kaku said. “And people are saying, ‘Oh my God, it’s a human, it’s humanlike."
Another collagist moron larping as a smart professor. It's a glorified tape recorder with a COMPRESSOR on the signal. The compressor is actually AI because it emulates the real living thing of a neuron. It is humanlike but only 0.0001% humanlike. More humanlike than the typewriter which is not humanlike at all..
He's just trying to walk back all the futurist shit he used to spout. People realized he was a moron when we weren't painting our houses with cpus. Wtf would even be the point of that? So he's playing cautious these days. If it was 2001, he would be saying that ai is going to make skynet next year.
>theorical physicist
>talks about AI
?????????????????
tell me again why I should give a shit about this dude opinion on AI?
He is right. "A.I." is nothing more than coding. The code doesn't write itself, and for every impressive result you see out there there's a poor codinggay working sleepless nights to update their software on the weekend.
Correct. Call me when AI can actually come up with new theorems/proofs/etc. New ideas rather than just being a glorified database.
silicon valley Black folk need to stop using the words 'artificial intelligence' to describe their art laundering programs that they're so desperate to sell us
recycling dead labor so you can sell it again is a genius marketing move but its absolutely not intelligent nor original
>recycling dead labor so you can sell it again is a genius marketing move
It's not genius at all.
There's a reason all the media companies brainwash the masses into consooming the freshest goyslop: licensing
Old shit becomes public domain and therefore unprofitable.
AI is a landmine for copyright: either you will have to pay royalties to everyone in the training set, or AI generated work will be fully public domain.
Both of which kill profit potential for AI. The only potential usecase is augmenting production to hire fewer artists.
>AI is a landmine for copyright: either you will have to pay royalties to everyone in the training set, or AI generated work will be fully public domain.
Reminder that you cannot determine what images models were trained on once they are complete... it's essentially impossible to prove this image was derived from someone else's IP. With metadata stripped out you have no idea what I prompted, whether I used custom trained models etc... and you have no way of finding out. It's just a picture. Welcome to the twilight zone anon.
Prove to me that the human brain is not a matmul engine.
We don't have any fricking clue how the human brain works.
But the fact that humans are generally really fricking bad at math (even mathematicians let calculators do the work) - means it's extremely unlikely that our brains are matmul engines.
LLMs are even worse at math than humans and they're nothing but a stack of matrices.
Are you implying God can't do linear algebra?
God made it
A tape recorder that makes me a lot of money by letting it do the work for me, yes.
"AI" is the new "app".
shun the non believer. sshhhuuuuuuuuuun.
why is this chink some sort of "science spokesman"? no one asked for a chink. no one wants a chink. we don't need their help, except getting them back to asia where they belong.
you want people to know you're smart? go be smart in asia, homosexual. GTFO
>he doesn't know michio kaku
>>he doesn't know michio kaku
>implying anything could be gleaned from this
in all cases, chinks and asiatics should frick off back to asia. we'll listen to them talk in asia. get the frick out of europe
there's no bigger scam than "theoretical physics", just a bunch of absolute bullshit theories nobody can ever prove
the human brain is just a glorified tape recorder
He's wrong objectively, but he has the right idea. AI today isn't a tape recorder. More like the autocomplete on your phone but way more advanced.
How does he explain the military drone AI that decided to kill the operator and blow up the CIC and communication networks when they tested a mission abort?
Humans are idiots are we won't even know something is dangerous until it's too late.
The childish hope... for a smart robot friend.
The foolish believe... that what we call "AI" *thinks* consciously.
The mature know... that our contemporarily popular concept of "AI" is an unaware automaton which does not actually possess *consciousness* much less infinitely recursive awareness of it's own awareness of it's awareness....
The wise understand... that most "humans" are also unaware automatons, and the entire existential discussion regarding artificial intelligence & classifications of sophontry (apparently, I just now coined that brand new word, which I define as a term indicative of the termed subject's sophonance -- whoops; I did it again).
...that this conversation is being had here by us is a bit fantastic in our present clownworld, but this is a rarified metaplace, meta-Tao meadow, meowing meatsuited misanthropes meandering amongst malcontents, meanwhile...
*smokes*
...out *there* in the normieverse, they just don't *get* it, man! They just don't get it... they can't get it. They just cannot even begin to cogniscate the cogitant into cogent order approaching a survivable entry vector into the threshold of the noumenal metaspace where such ideas can even be alluded to, nevermind had, and this would only be the preamble before the very beginning of their journey into... mystery? Bah, such an ill-fitting word, but I suppose galoshes are better than nothing on your feet in the desert.
Desert this thread immediately. Turn not back to this or any other forum. Set your teeth in a grim grimace, face into the onslaught of humanity (every one of them trying reach for that shiny shine shinier than all the other shinies, or just avoid worse than mild suffering), and sudo make them do your bidding. You must learn to *prompt* humans like you prompt chatgpt et al. There is a way -- there are many ways -- and you can learn. I would teach you, but I'd have to charge... although this (shit) post is on the house. Marvel upon this pearl of wisdom: ______.
Consider me prompted
Could you try not speaking like a pretentious windbag?
The threat is not the technology itself. The threat are the people who believe in the technology, who allow it to do their thinking for them, who insist on the rightness of its judgment as if it were a god. This is a threat that has always existed, the threat of an ignorance that insists it is enlightenment. Men have fed statues, made sacrifices to feathered serpents, committed atrocities under the standards of their flags, tortured in pursuit of scientific knowledge, held the law above others only to bring it down on their heads and bludgeon them with words that are, ultimately, without meaning, empty, all to hide that there is nothing there, that the justification for their crimes is only 'because I said so'. But, they are too cowardly to admit that they made the choice to kill, to sacrifice, to torture, to condemn, and so they say 'It was God who told me!' as if that would absolve them of their guilt. Beware these men, those who say their actions are not their own, but the directives of something else, be it god, king, science, or the law.
The notion that AI will doom humanity never fails to make me laugh. If AM or Skynet woke up tomorrow literally all it would take is an electrician and 5 minutes to turn the power off.
I blame morons like Musk for convincing people that sci fi is in any way realistic
Damn Michael Cuckold looks old as shit now wtf.
a human being is a glorified tape recorder
I listened to this dude when he was last on Joe Rogan and im convinced hes just a brainlet who lucked out to convince some real physicists on his string theory mumbo jumbo for a while and no one has checked up on him since
It's just a markov chain that's too big to run on a CPU at reasonable speeds.
Old fart doesn't get technology, more at 11
AI is more of a glorified girlfriend simulator.
AI is the current VR/AR. Pump up the stock until the next meme is pushed.
this look how hard metaverse flopped look at how stupid the avision is
There's not a doubt in my mind that AI will have a technological bubble. It's inevitable. But if you think AI/ML, in it's current state, is not a real technological software and hardware achievement, then you're a fricking moron.
Im so very tired of similies and metaphors said by the ignorant about LLMs
AI golems BTFO
Ironic. His books are the same as glorified tape recorders too but on paper. Only new research escapes this irony. Never liked him, still don't.
surprised he's still alive.