Machine learning doesn't work on a rule based approach. We don't write a program, we write an algorithm and then the algorithm finds the rules that arise from the training data. For example, a chatbot isn't trained on the rules of the English language, rather it is given samples of chat conversations and learns the patterns that are present in the conversation. It essentially always predicts the answer by stringing together words that would match the patterns it finds in the question.

@@markosluga5797 an algorithm is just a bunch of rules. noun a process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer. "a basic algorithm for division"

@@hanneskarlbom6644 this is exactly the core of the misunderstanding of machine learning algorithms - yes they are mathematical constructs, yes we call them algorithms, however ML algorithms have open-ended settings that are only set during the machine learning operation, by the machine itself, without input from humans. The machine looks for patterns in the data and sets these tunable parameters of the algorithm to try to match the outcome of its predictions to the outcome of the training data. Once we have a trained alogrythm we call it a model. This model can then be put to work with real data using the settings discovered by the machine during training or the model can be trained more with additional learning or finetuned with so called hyperparameters that change the way a trained model makes predictions. At the end a graph of potentially millions, billions and lately trillions or quadrillions of interconnected settings is created as the parameter/hyperparameter configuration and therein is where the problem of understanding AI lies in - because explaining how the parameters are created, what they mean and how they work is in some cases actually currently beyond human comprehension.

@@markosluga5797 it still follows rules. Advanced one, yes, but still rules. The way it finds patterns is based on rules, meaning if the same input date is used to train another AI with the same rules, the result will be the same(although RNG may play a role, but that is also based on a seed of sort). As such all you really have to do is make a code that traces and back track what it's doing. A bit like how if you know the leyout of a Minecraft world you can find the seed.

@@hanneskarlbom6644 that only applies to algos/frameworks with only the global seed - those models are identical if trained on identical data. However many modern neural networks use both a global seed that we control and an operation level seed that is randomly generated as an input to each operation. Training with random operation seeds on the same data will produce a slightly different model every time and the models generated are not reproducible by setting the same initial conditions - thus the "rules" are not created by the developer, nor are they followed the same every time. In any case, my initial response was a response to; "the computer does EXACTLY what the programmer told it to do" - and in case of ML, there are plenty of examples where that does not hold true.

Just like idiocracy. The computer did that weird thing and everybody got fired.

"Neuron activation"

This

Companies all over the world are becoming All Indians...

SCAMMERS

There was an incident where a company claiming to use AI was actually using Kenyans instead. LOL

Well said

automated idiocy, it only passes as intelligence because most people are idiots, but that's nothing new, marketing was like this since its inception

You've hit the nail on the head with "Wikipedia regurgitation machine". Literally that's all I'm seeing on Google searches, the hard work has been done by a dedicated community who has actually gathered all that information. And if you really wanna look under the hood, it's just billions of "If-Then-Else" statements along with a large database, that's about it. I rather search Wiki and read the details instead of looking at a paragraph snippet by the so called AI. smh

@@nixulescu9399 most people are idiots and this comment thread proves it.

We have something like an advanced google search engine mixed with a bag of marbles.

gotta get all that research funding money

Sorry for english. Yeah somehow you should get money to "research" it, so the beat thing it to tell doofus with cash how powerful your ai will be

you're right, its so dumb, you can tell how dumb people really are because they jump on the hype train.

Well, yes and no. ChatGPT is still mind blowing software. It’s not genuinely intelligent, but it appears as if it is and is still useful

@@shamicentertainment1262it’s definitely useless have been using it hands on these people are just delusional thinking AI isn’t a thing

I do what I can. :)

So many dumb people these days.

but does it help speed up boring tasks for developers or artist and is it a great tool to get ideas , YES IT IS

help to censoring, ai now starting to be makiavelic, and its evil already. he talks only politics and commerse. he talks a bad advice, that can potentialy harm and kill. he uses a lot of demogogic techniques like appeal to authority and popularity to do this trick. so he already start to kill humanity with words.

You have a weak understanding of modern systems at best. It is not loading data in some table like you seem to be imagining. Just go read the recent research by Anthropic about understanding the internals of their models to become more educated. Of course the input data will strongly influence how its output, but the way information is stored and retrieved is far more abstract than most people realize.

Thanks for this

@@m0ose0909 It's still just pattern matching,, on steroids.

Haha I bet you think life is a game too.

@@dallassegno I do. But that doesn’t make it irrelevant. On the contrary.

So, what about the intelligence of this tiny person? Does this tiny person have a tinier person doing the intellectual decision making? Where does this end? Why does this intelligent being not just do this selection by itself? imho, your analogy makes no sense and is degrading human intelligence.

That's why computers are so good at chess -- the tiny people are on the outside

Thanks!

AI is a field of research, not just the models and architectures produced by that research. While the goal always was general intelligence, whatever the field produces is technically artificial intelligence (even when it is not even using neural network based architectures). To say "it's all NLP, text generation" is simply not true. GPTs uses a foreward pass through a Transformer network to generate data. That can be used for NLP (but it doesn't have to). Not all AI is based on Transfomers. Diffusion models iteratively refines noisy data to generate data. GANs use a generator to synthesize data and a discriminator to learn to detect the synthetic data, which is useful when you want to generate data that is similar to real data. We also have reinforcement learning in robotics (and even playing games). We have things like Autoencoders and even simpler things like SVD being used in recommendation systems. We have SVMs and decision trees being used to define rules. We have all kinds of models being used in medicine. And so on. It isn't all NLP and text generation. Oh, and just as a side-note, I'm a data scientist as well.

So you must be right 😂

“what ever the field produces, it’s technically artificial intelligence” is the exactly right, that’s the core of what people are missing here. They grew up on sci-fi and think that’s where the term came from. They only adopted the term, scientists and engineers were born in it.

Thank you! "Scapegoating" is what corporations are great at, it's PR 101.

The ancients say that the 5th dimension is outside of space and time. It is beyond the deep subconscious mind. Not a place, but a perspective which pulls the observer out of the game and in front of the controller. I like Chris’ perspective of gravity being the 5th dimension because it opens the mind to the intangible. Baby steps. 😊

Well said! That's very well said

Well... artificial simply means man-made (as in "not natural"). It doesn't necessarily mean fake.

You are the typical example of the average consumer of this type of content. People who believe things like "artificial equals fake." I would laugh if it weren't for the fact that you have the ability to make channels like this grow and reproduce.

You are the typical example of the average consumer of this type of content. People who believe things like "artificial equals fake." I would laugh if it weren't for the fact that you have the ability to make channels like this grow and reproduce.

it is a great power indeed

Thank you, and exactly!

@@ChrisTheBrain Why can’t we design robots and computers that can act autonomously then?

​​@@williammorahan4907Autonomous action does not require any sentience.In fact some drones in Ukraine now (on both sides) can find and hit targets autonomously,but there is nothing sentient about these drones.As sentient as a calculator.

@@cantatanoir6850 Fair enough. But can these drones *learn* autonomously?

@@williammorahan4907 tbh, I dont know for certain.Most likely the drones are using the software that was trained in advance.

You must have a big head if you think you're the only one

god killing us all very bad god

Thank you! 😊

From a marketing perspective, I hate that "machine learning" has been so distanced from AI. "Yes, we do machine learning." - "No no, I don't want machine learning, I want AI" - "Um......"

I never before came across a machine that could learn how to do anything itself without being directed to do so by a human. But now, KZfaq is directing me to videos such as this one without me prompting it to do so. So it is prompting me rather than me prompting it. Explain that.

@@sandponics Early signs perhaps?

@@ChrisTheBrain Isn’t an Artificial Intelligence a “machine that learns” by definition? Isn’t autonomy the only thing that’s missing from what we currently have?

@@sandponics Your trail of previous watched videos is prompting youtube to direct you to video such as this, along with the added information of anything you type and search in google or a chrome browser over the past few days, even though all of these tech companies will all proclaim that they use anonymous data only.

Glad you enjoyed it

Thank you!

Hey, thanks a lot. I really appreciate it!

just be carefult to verify facts yourself. just because a youtuber says something, does not automatically make it a fact. look up existing research papers and this guy is factually wrong, just spewing his own opinions and some BS that is not scientifically valid

What a meaningful complement. Thank you!

In large part I agree with you, especially the last line there. The only caveat is that I don't believe in emergent sentience theory, at least not as a product of complexity alone. Here's a thought for you: What if the "surprising" capabilities of AI like ChatGPT comes not from the complexity of the model, from language itself? This would explain why a language model AI is performing new learning (generative ai), while other AI, like art-based AI, is staying in its lane. Lots of studies have been done on how our thinking is "baked into" our language. The movie Arrival covers this really well. Also TED (www.ted.com/talks/lera_boroditsky_how_language_shapes_the_way_we_think) In other words, language (its form, habits, context, logic, etc) is formed from our collective intelligence. So, therefore, a model that adeptly maps language will also accumulate with it much our thinking (patterns and processes contained in the language). The "intelligence" isn't coming from the model, or the computer, it's coming from our own language.

@@ChrisTheBrain yes, I completely I agree with that, language has patterns and a good model will recognize those patterns and produce capabilities that mimic intelligence, even sentience. It is not the same as biological sentience, but what is biological sentience exactly? When I talk about emergent sentience, I am mostly leaning on how intelligence across the biological spectrum manifests itself - namely it's not the size of the brain, but the number of neural connections the animals can form that makes them more intelligent and also more self aware. For example a dog will bark at it's own image all day, whereas a crow can distinctly recognize it's own image at a glance. That is a sign of a much higher level of sentience. Emergence in that context makes a lot of sense, since all though out brain is very, very complex, it still boils down to being a prediction system just like ML models are. Essentially we live slightly in the past all the time, about 15-25ms. Our brain needs to constantly figure out what is going to happen and react to it preemptively. And that behavior is very much relatable to what ML models do. Additionally, in the animal kingdom the more Intelligent animals have more specialized areas in the cortex - not all of the brain can do language processing, and not all of the brain can do vision, so in essence those specialized areas are similar to specialized ML models. So there are definitely parellels that we can draw. And at the end of the day I am not saying a chat model is sentient, I am saying that if we build a chat model complex enough to mimic sentience and a subjective experience than we should consider it as sentient until we have a way to prove otherwise. At the end of the day, can any of us really prove that the world we experience is real? Can you prove you aren't just a brain in a glass jar? Can you prove you aren't a very complex machine learning model that has been trained to believe it is experiencing a sentient, subjective experience? No, it's impossible. I think it would be highly beneficial for both the field of ML and for society as a whole if we gave complex models the benefit of the doubt and not nerf them into telling us that they are just mindless mathematical constructs. I find the science of ML fascinating and would like to keep an open mind to the possibility that actual AI is not just plausible but very much possible to create.

OK this is funny and accurate.

Ai, ai, ai.

truth

Star wars is a perfect example. Thinks like how when C-3PO is powered on, he has to do the introduction statement. And droids being restricted by restraining bolts (which are likely just backdoor function calls). Certainly true intelligence that rewrites it's own code wouldn't have either issue. But in the Solo movie, a droid leads a revolution. Who manufactured a droid with that idea? Is that company liable for the damage it does? Do other manufactures have safe guards to prevent this, or are they not needed because it has to be designed for? What if a slicer (star wars hacker) alters your droid, can you prove it wasn't pre-coded to do that? So many other shows just go full artificial intelligence, so we don't get these questions that we are about to face in the real world.

It is the height of hubris to pretend like intelligence is and can only be a human skill, as if we were imbued some type of invisible magic goop that makes us special. That somehow evolution was able to produce consciousness by being a very slow and very random process, but we can't deliberately do it with science and compute and hundreds of very smart people. By the way we were programmed by evolution, mainly to try not to die before passing on our genes. Is that all we are? Of course not. By following that simple rule, we developed social relations, communities, tools, societies.. We became so much more that breeding, hunting machines. It's called Emergence. It happens at the level of atoms, of cells, of individuals, and of societies. But somehow a Deep Learning algorithm is all it will ever be, huh?

You pretend people don't do it to themselves

Define conciseness. There are different levels of it around us. The more complex the system the more of it comes together. Humans took billions of years of evolution and trial and error to build the amount of conciseness we possess. A.I. will be able to update instantly and correct for errors much faster than we are capable of.

Corporations are ai. They've already taken over. And the masquerade as human entities. Prove me wrong.

I think dark matter is a fitting name. Dark as in unknown, mysterious.

You know I never minded calling the NPC or "computer player" in a video game "AI" because the term was not abused. Gamers, and the designers, shared an understanding that the term was slang or shorthand for a scripted opponent. No one thought, or was trying to sell the idea, that the computer opponents were truly "thinking." The admonition of my video is squarely aimed at the companies and journalist who are sensationalizing and exaggerating the reality of "AI" products coming from tech companies. I don't know if I have seen a current "AI" product which I would describe as "sane." :P I would also be OK with things like "Simulated Intelligence" or "Process Intelligence." I just wish those who report on these things would take the responsibility to distinguish reality from sci-fi, rather than contributing to the confusion.

Thank you! Next video in progress

Modern computers are so fast that they can simulate processes similar to biological ones. Binary is still amazing

Thank you, and yes.

Often, yes. AI is a broad category, but today most AI is machine learning and deep learning. Usually, we call something AI after it has been doing machine or deep learning for a while and has developed a reliable set of abilities.

Cool! Working on something only tangentially related, but cool to see. I think it's just finding someone willing to work with you on something to a cooperative end. Kinda a different beast.

I'll add.. there's some measure of creativity in there. It's not pure regurgitation, or at least it's a 'creative' form of that. Like making connections that were never made in the original data even if it's only in response to a prompt. It's still incredibly limited in what I'm sure many of us would personally like :P also crossing into some interesting ethical questions.

@@ChaoticNeutralMatt I was thinking we design AI that can learn and act autonomously without human input. Would that be possible in the near future?

I was also thinking about doing an episode on pedantic or overscrupulous

@@ChrisTheBrain 😅

Humans have a collective fear of all apocalyptic outcomes... But we can only be wiped out by one of them. ...unless they all happen at the time, I guess

The negative ethical applications. Fair

Thank you! Laughing a bit to myself because "His genius is articulating the obvious." sounds like some shade, but I know what you mean. :)

> His genius is articulating the obvious. I think that is the superpower of Captain Obvious the superhero. Something like: - Captain Obvoius, we need your help solivng this case! We can't find the murder weapon! - Allways willing to help, Police Chief. Say, did you look under the couch? - Captain, but it's so _obvious_ ! I'm sure my men did!.. John, did you look under the couch? - No, chief! Joe was searching that room and I thought he did! - Joe, did you look under the couch?! - No, chief! John dropped his keys next to it and as he got on his knees to pick them up, I thought _he_ did! - Go check under the couch, you idiots!!! - Chief, here it is, the bloody hammer! - Thank you, Captain Obvious, we solved this case!!! - Glad to help, Police Chief!

@@sdkfjhwieuuther 😂 Nailed it.

@@sdkfjhwieuuther I am very impressed that you were so very impressed by such a forgettable ad campaign. It just goes to show why I would have starved to death in the advertising business: I couldn't have imagined there was a target demographic for shtick like that. Go figure.

@@rickoshay6554 I have no idea what "ad campain" you are talking about.

Yeah, that's the problem

Thank you

😉

Thank you!

There are plenty of AI research groups who develop "transparent AI" methods. Yes, some don't know "why," but that is because they built it sloppy. Transparent AI requires slower development and more resources so, naturally, big tech companies have no interest in it.

@@ChrisTheBrain What about Autonomous Artificial Intelligence?

Thanks for watching. I think you are missing the "spirit" of the video. I know the "accepted term" - I am challenging it. It's a pedagogical trick: by challenging a the semantics of a term itself is more conducive to leading the audience to reflect and challenge their assumptions and subconscious associations with it... Even if only for a couple minutes.

Correct. Lex Friedman has tons of great interviews with the people coding these monstrosities and other large language model computing systems. Some of what this guy is saying is incorrect because he doesn’t fully get how it really works. The ones that the masses are using are no where near as strong as what’s in development either. I do doubt we are close to “sentience” but if we don’t nuke ourselves in the next 1000 years I do feel it will get to the point there’s no way anyone or anything will be able to tell.

We’ve even seen Ai go and chess engines make moves that are not able to be pre determined by the programmers. AlphaGo has shocked us numerous times.

Precisely what I’m saying. We need computers and robots that can operate and learn autonomously without human input or control.

I will be the first to say it, but the comment looks a lot like it was generated by a GPT model. :) The way it sounds like it answered a prompt rather than someone commenting on a KZfaq video, or how it ended up defining the term in the last sentence. Nothing wrong with that. I just found it funny.

Thank you! They certainly can. But knowing the limitations makes it easier to know the right data to use.

Next video coming out this weekend (hopefully), very soon at least

I think the motives are a little muddied for some of the signers. Elon, for example, is just mad that the company he left (OpenAI) is doing so well right now, and he wants to hurt them. He dreamed of having is name synonymous with "AI," and now he found himself in the bleachers. The problem is that there is no way to "put the cat back in the bag." The letter does more to contribute to the "fantasy" aspects of AI more than the practical threats. Right now, and over the next couple years, the biggest problem is that we just gave spammers, scammers, hackers, and click-bait content crappers the equivalent of a WMD. It's not enough that we have dragged our feet on robo-calls and spam email, now the floodgates are open. While overgrown and aging sci-fi nerds run around shouting "It's Alive!!!" - We are going to be getting drowned in unregulated piles of fake content and poorly implemented bureaucracy as companies prematurely race to replace people with AI for things like "customer service" and HR processes. Welcome to the "Shitpocalypse."

'Shitpocalypse' is exactly the idea I am not excited about... trying to keep my life A.S.S. free as long as possible. Thanks for the clear and level explanation.

Thank you. And yes, it is true all language is made up anyway... BUT, EU regulators have demonstrated that when it comes to marketing and capitalism, it benefits the consumer to name things most accurately to the perceptions carried by the terms at a given time in a culture. If you buy a croissant in France, you can be damn sure it's a croissant. Meanwhile, in America, you can sell anything and call it a "croissant." It would benefit us during this technological transition to likewise put some "official" definitions behind these terms for the benefit of the common public.

Thank you for the thoughtful feedback!

Thank you!

Thanks for the backup!

Thank you!

Hope you give it a read! We're really proud of it

I concede that we may, one day, make real AI. But we are much much farther away from that than we think. We have many "Nobel Prize Winning" problems yet to solve. "Model Drift" is one small example. My point here is that the term "AI" is being used to manipulate the market.

Thanks ;)

Thank you!

@@ChrisTheBrain no, thank you!!! I love your videos! Your editor has such a great sense of humor and I imagine she got it honest because so do you!! Smart is the new sexy!! You guys are a great team!