Future Computers Will Be Radically Different (Analog Computing) 

From USA?

There’s a reason he has 11, 000, 000, 000 subscribers after all 😏

@@alex.g7317 Wow, that's more than the population of Earth. Where does he find all those subscribers??

@@unstable-horse Mars

@@unstable-horse omg, lol 😂. That was a typo! I meant 11, 000, 000!

Let's hope these new analog chips can solve our GPU shortage problem 😂

In general when doing machine learning you are only using the CUDA cores of a graphics card so the wattage never gets close to its maximum. A lot of the processing units are simply not being used, for example shaders and 3D processing units. For example on my GTX 1080 I sit between 60-90w out of 200w when doing Pytorch machine learning. So I think 100w out of a maximum effect of 250w seems reasonable.

you can underclock GPU's, thats what they do in cryptomining to improve their profit margins, depending on the chip they can operate effeciently at a fraction of their nominal power

man, I'm old enough to remember when a $1200 card was considered EX PENS >IVE< And not...'going price'.

@@AC3handle 1200 wont buy you enough power for a decent DL rig either. An RTX 3090 goes for ~$3000 USD

I know this is an old comment, but I figured I'd add that as far as physical packaging goes, nothing stops us from putting one of these next to a conventional CPU. Cooling it would be the hard part as the temperature would swing the outputs by introducing noise. Might be better as an M.2 PCIE device.

@@DigitalJedi Its incredibly unlikely this will ever expand into the home. These would likely be built entirely different from traditional computers.

@goldenhate6649 As we've seen they can be build on NAND processes already, which are widely adopted by consumer electronics. The use case provided of low-power wake-word and condition detection seems like a great application if they can find the right product in the consumer space.

Great recommendation. Thanks.

A Logical Calculus of the Ideas Immanent in Nervous Activity

My first thought when he pulled out the analog computer was "Hey that looks like a modular synth!"

Witness Audio Modeling (search for it on RU-vid).

Music has always been an AI task

As a semi-professional music producer with almost half a decade of working with professional musicians I would agree - and this is mainly because people feel a lack of “soul” in music. Those small human errors that we’ve spent decades trying to get rid of with Autotune, Drum machines, Sequencers, digital synthesiser and digital samplers (the last two CAN create sounds that will always come out the same way as long as the input stays the same - however there are exemptions). This is probably something the people I know in the music industry refer to as “The generation-rule”, in brief the music today is a result of what our parents and grandparents heard combined with new technologies and pop culture. - If you’re interested in music and maybe want to stay ahead of the game look it up. Some refer to it as the “30 year rule” as well.

@@p1CM AI has no tasks

At the expense of accuracy?

Well, semi analogue. Don't forget the bias (voltage drop) before you get current amplification. Also, to say that analogue computers are more power efficient that digital is pretty hard to back up. A $2 microcontroller can run on a few mA for the desired task, then sleep on uA. You'll need at least 5mA for an analogue computer to start with and you can't make it sleep.

@@JKPhotoNZ Great point. And with current Nano transistor technology, That efficiency (along with raw power) is going far beyond what a true analogue system could produce.

Yeah but then you'll never know at which zone is it on? Because amplification happens differently for different input parameters. And not all transistors from the same batch would perform the same, i.e. it will lack repeatability (as Derek mentioned).

the insoluable (even in theory) problems of analog are noise and signal integrity, which is why he didnt even mention them. This channel has gone to poop honestly.

To be fair, many of the tools we use are analog. We just don't call them "analog computers"... even though, they kind of are.

Exactly. Museums is where analog computers belong.

@@rogerphelps9939 The words of someone who knows nothing but his own little world. And he is content with it. Honestly, i'm jealous. For real, stay that way or life will get an awful lot harder. I would give everything i have to aquire such luxury.

True AI is going to be the end of us. Why would you want that?

@@teru797 true AI wont be possible for the next 200 years and by then if humanity kept on living how they are we aint gonna survive anyways

@@teru797 it would still take quantum computers to be able to have the memory necessary to run

@@teru797 cool

@@teru797 we are already the end of us

Thus the temperature coefficient is very important for recent precision devices. And a high accuracy low ppm resistor is expensive, which is one of the reasons why it costs so much for high-end electronics instruments.

I was going to comment that one disadvantage of analog computers is keeping them calibrated. If you want a precise amount of 'voltage' or movement to represent a real-world value, you have to keep it calibrated. Older mechanical ones had wear/tear, electronic ones have issues as well.

Ah!? But what if the resistors were warming up, digitally?

@@stefangriffin2688 you can't have a digital resistor

@@stefangriffin2688 yeah digital signals works with gates - on or off

You must be really healthy because you dont even look close to 70! :D

@@magma5267 Thanks for the heads up. I've got a good women, 6 children, 8 grandchildren and a recently placed stent in my heart that keeps me going :)

@@TerryMurrayTalks Good for you my man! I am still 20 and don't know what to do in life :(

Yep your 70 mate

@@vedkorla300 You have plenty of road left to travel, follow what you love, enjoy the journey, don't let the bumps in the road stop you and if you can get a soul mate to share it with you it will all be good.

That's pretty much where things have always been. Using basic building blocks that do specific functions. A linear voltage regulator has the job of maintaining a constant output voltage for a given set of current levels and different input voltages. You can buy an Opamp and using resistors to make a function called a schmitt trigger. Or you might just buy a schmitt trigger from Texas Instruments and put it onto a board with less board space consumed. Or a schmitt trigger might be embedded for free in certain other ICs (intergrated circuit). The major computing engines I have seen so far have been effectively GPUs, CPUs, and FPGAs. Xilinx &Altera (now Intel) have specialized in making FPGAs. An FPGA's basic internal components are logic elements with flipflops with reset, aysnc reset inputs, 4-input look up table, etc. Cascade these to make larger units like a multiplexer, floating point arthimetic unit, etc. Its programmable so you can effectively emulate a worse performing specialized CPU. A CPU is still more effcient at doing CPU type functions. A GPU does specific stuff as well. The idea of doing analog computations honestly just sounds like another building block to add into a complex system. Only there simply hasn't been a large enough demand to require the generation of specialized hardware like what was described in this video. That one start-up sounds like its developing a chip that will do a series of very specific functions and will need to be integrated into a large systems to accomlish a specific task.

Well that have sort of always been the case. I don´t know what was the first accelerators, but one of the fairly early once was the FPU. We nu just take it for granted. Sprite accelerator was also fairly early. Then graphics accelerators. Then video decoder/encoder Then MMU accelerators Then 3D accelerators. Them SIMD accelertors. Then T&L accelerators Then physics accelertors Then raytracing accelerators Then deep learning accelerators.

ASIC devices have existed for decades...

@@LundBrandon they will soon become obsolete though due to reconfigurable computing devices line FPGAs

@@calculator4482 FPGAs have also been around for decades, plus they draw more power. I'm a computer engineering student right now currently designing a CPU to be synthesized onto an FPGA. I'm not dumb.

Op amps with their offset voltages and input bias currents leading to inaccuracy. Sounds like a nightmare. Constant recalibration required?

Absolutely.

@@rogerphelps9939 just calculating the uncertainty at each step sounds like a nightmare.

but you can mix with digital control for adjusting and decision

I think Nasa (or was it still Naca at the time?) was able to simulate flight caracteristics with analog circuits too. I'm thrilled to see this tech coming back !

Underrated comment!

Absolutely love this comment! This has been on my mind for at least an hour now, the point you make is intriguing and a bit haunting, thanks for that!

Fkin right on!!

Very interesting.

Underrated comment. Then in Eternals style we will have to reprogram the memories of our servants every now and then.

Absolutely. Love the way he covers the concept for everyone. Those who don't know in depth about it can still go away with a sort of basic understanding. And those who do understand it in depth will enjoy discovering new areas of invention that they can further explore. Looking forward to reading some papers on using analog computing in neural network applications

I'm smart too.

Same here. But i didnt go to high school and mylife became a total mess and i havent graduated whatsoever :)

“nerd-sniped” lmao I feel exactly the same and here I was thinking I was way ahead of the curve on alternate computing 😂 jokes on me

Damn you're old! Glad you're still kicking around

@@Ozhull he’s only like 75 chill

Lol

Woo fresno!

I graduated Fresno State in 1993 with a BS in CS (after a stint in the military). I even took an artificial intelligence class. I still have my perceptron book.

🇨🇦🏳‍🌈

Its amazing how fast all of this is evolving. Looking at this, and comparing it to facial recognition software in simple phone apps we have now really shows how much all of this has influenced what kids and teens easily use today.

I didn't catch the part where he quit talking about analog systems though when he went to the logic systems being used for matrix operations, because that was digital. There may have been analog inputs into the system, but there's an A to D conversion, and everything he showed at the end was strictly digital, so a bit misleading there. Current systems for AI are digital.

@@raijuko yes but i still believe in God . And i am a fan of science experimenting and inventing

Yes, indeed. Something I find fascinating are recurrent networks, where some neurons feed back into the network which would allow some information to be saved from one image to the next one. This would allow the AI to process things that change in time, like music and video. For example, if you're tracking a subject with a camera and he turns around, a recurrent AI would be able to continue tracking the subject.

Yea in the end to some of it is kind of voodoo black magic thought i mean they call them black boxes for a reason.

speak in english pls

The video seems to contain some quite biased infos though. The top 5 error rate of humans of 5,1% is of course not accurate. If human beings were that bad, we accordingly would have much, much higher car accident rates. Those kind of inaccurate percentages come from statistics based on captchas. And several conditions do distort the results there. - human users often don't bring up the needed concentration and attention to solve captchas as they actually could. In fact they are angered by them and often times just click quickly through them. In traffic while driving a car human beings are much higher on alert and do tremendously less mistakes. Here human beings still beat autonomous drive computers by several orders of magnitudes, measured in car accidents per million driving hours. - the captchas often do not meet the actual human perception. The captcha images are often unclean, got low resolution and distortions. In the real world humans perceive much higher quality from their surroundings than some crippled captchas. A more clear image increases the recognition dramatically. It is really crucial in educational videos from whom and from where you take your numbers. Science is not always that objective as we are told, especially when corporations are funding them financially with own interests. Other than that the video is quite interesting. I also wished though that many common misconceptions would have been cleared up. For example many people still believe, that computers would work like human brains. This is plain nonsense, mostly spread by science fiction. The brain still does pose big mysteries to us especially "the big problem of consciousness".

It is still kind of voodoo or black magic. While the overall working mechanisms are well known and the output can be estimated based on the input, how the neural network exactly reaches to the answer is nearly impossible to inspect because of the sheer amount of variables. In essence, you feed a black box with something and you can expect it to give you a particular answer with some confidence, but no-one has any damn idea what exactly happens inside the black box.

Aot of the actually do! You can even (kind of) help your brain waves to synchronise with the oscillations, it's not by brute force (you have to play along or it doesn't work very well) but it can greatly help for sleeping, learning and related stuff. Reminds me of old hardware synthts, where you had to connect each of the synth parts with cables. But tat gave you an amazing amount of flexibility! BUT, no one cared to write the configurations down... That was te funniest and most awful part at the same time...

Binary computers have to deal with interference too, that's handled by error correction. Error correction is already baked into the infrastructure of every digital component, to the point where we don't realize it's there. They suggested one method of error correction in the video, and they're probably not even scratching the surface of what's possible.

i think the crux is the medium, the AI models brain which is so good at re-building itself, and it uses electron and chemicals to convey information. our transistors are too limited to mimic that interaction. for example the new 3w chip needs to be custom made for each model if i got it right.

@@introprospector Yes, but with binary error correcting is simpler as interference isn't as much of a burden on the architecture. When the job is to check if there is, or isn't voltage it is a lot less complex than checking 8 different voltage thresholds.

@@fltfathin This is called an ASIC (Application specific circuit) the computer is pretty much just sent to the landfill after it's outlived it's usefulness instead of having the ability to be repurposed. Which is another concern about where computing in general is heading as PCBs use less and less semi-precoius, or precious metals there is less incentive to recycle.

@@dust7962 Yeah but I put my mobo in the case first and then the radiator wouldn't fit, so I had to take it out and install the radiator first. It was really annoying. I didn't watch this video, but I'm assuming this is what he was talking about.

I have never done much/extremely deep research on a topic, but this seems very interesting.

id like your comment to not be burried.

0s and 1s . high and lows. voltage and no voltage (digital representation of numbers); have absolutely nothing to do with brain neurons. it's complete BS . For all we truly know about the brain there could be a near unless amount of information with every firing of a neuron. We have no idea what format conciseness information is in and likely never well as "in time" humans.

@@noblenessdee6151 Engineers : well I will approximate every thing a neuron is saying, to just two numbers. 🙂...

@@AayushPatel-gc3fw words are the limitations, not numbers, all words can be expressed in code, not all humanity can be expressed in words.

They already use crap like this far too often. This isn't something use for for a differential equations course. Maybe it would be ok for a circuits course or even Computer Science course focused solely on analog computers. In math, just give us the numbers and the logic... Don't waste are time with this stuff.

@@JackFalltrades I am an engineering student.

@@JackFalltrades I'm not calling letting students know of use-cases for things crap. I'm calling taking up class time that is meant for teaching students the logic of how to solve differential equations (because that is the class the original poster said it would be good for) and instead using that class time to teach something that only a tiny fraction of the class would ever use.

it is used in control systems

@@Elrog3 He never said it should be used in a differential equations course. You just sound like the type of students that go to university and ask which courses they need in order to get a high salary position in industry.

He lied.

@@rogerphelps9939 the Space Shuttle did indeed use analog computing for some of its flight control systems. the Space Shuttle used a hybrid digital/analog system for flight controls. Most of the high-level control logic was handled by digital computers, but critical low-level control functions were performed using analog circuits. The analog components were able to process sensor inputs and produce control outputs much faster - on the order of microseconds - compared to even the fastest digital computers of the era, which took milliseconds. This speed was essential for stability during flight.

I can imagine it too be also just more fault tolerant. Discrete = hard, continous = easy. An overflow in digital can literally crash a whole system. In analog there is more room for error.

ok

but it isn't, he literally said he's going to skip back propagation (which is how models are trained nowadays)

i am a layman I did not understand a bit of it, pun intended

You are still strugling with interconects from one side to the other

@@ravener96 With many ML algorithms you can split problems into multiple sub-problems for different networks to handle. I wonder if developing that area of ML would be helpful to make effective analog systems? For an example, in image processing a pixel at the top left of the image has little interaction with a pixel in the bottom right of the image compared to nearby pixels. If you wait to compare them until multiple layers later, it speeds up processing the image and allows for algorithms to become more adept at finding sub-patterns in the image.

@@Zeuskabob1 Depends on what kind of images processing the neural network is doing, if the computer wants to identify a face in a person maybe it doesn't need to process all pixels once it has processed all the pixels near the face, but in some cases distant pixels can indeed be correlated, like the images from a camera in an autonomous car identifying the white lines of a street, where it could be 99% sure it is a straight line but the corner pixels clearly indicates that is curve line.

@@ravener96 Yeah, but interconnects can be designed around with clever architecture to an extent. It's still quite interesting.

Flash cells are micron scale while the AI accelerators doing integer operation are built with the latest 4 nm technology. And floating gates have really limit life compared to pure logic circuit.

You can do everything digitally that an analog system can do and more. An example is resampling in order to change the frequency scale of a recording. This can be done in real time using digital methods, not so much for analog methods.

@@rogerphelps9939 Depends on what you’re doing and what’s important to you. Analog synths are great for experimenting with the knobs and patch bay (if available) and learning what exactly each change has on the overall waveforms. They’re really great for learning what exactly you’re doing and what you’re getting as a result. Yeah, there are software synths meant to emulate hardware knobs and a patch bay, but I haven’t found clicking through all that as valuable as plugging and experimenting yourself. That stuff really depends on the person, though. What doesn’t depend on the person, and is arguably more important, is the fact that aliasing can end up being a problem on digital synths. When you start doing some crazy cross modulation between sources and/or you’re dealing with lots of harmonics, if the processor can’t keep up, your sound will suffer. Same with super high frequencies. Depends on the synth, of course, but analog synths can tend to have a warmer, purer sound to them as well, because you don’t have to emulate all those harmonics. It really comes down to the same arguments being made here regarding analog computers: there’s no processor overhead needed to create some very complex shapes, and to do so perfectly accurately, on analog. I use both types of synths, as lots of people do, and I would never say that one somehow makes the other unnecessary. Hell, there are hybrid synths that give a mostly analog signal path while allowing for, say, a digital sample and hold circuit and the ability to save certain parameters. People make those kinds of things for a reason, you know?

Pythagoras discovered math with music I think right? Really like your comment

@@rogerphelps9939 Difference is that you need billions to trillions of transistors to do digitally what can be done using tens to hundreds of transistors analogously.

@@RAndrewNeal Wrong. The errors arising from component tolerances, noise and temperature dependent offsets make anything complicated pretty much impossible in analog. Transistors in digital processors are extremely cheap. Provided you have good DACs and ADCs you can do anything to whatever precision you need in digital.

The problem is machine learning is still not capable of being used commercially in general environments (e.g. security cameras) because they can't handle unpredictable situations. The brute force method of AI is still the only solution for general environments (e.g. self driving cars)

@@chrisfuller1268 people also suffer from the brute force nature of processing information, aka confirmation bias. Thankfully we can take steps to correct this, but many people lack the tools and mindset to make these self corrections.

@@riskyraccoon Yes, humans are flawed, but we are capable of recognizing objects no matter what else is in our field of view. This is a task machine learning will never be able to solve in 100% of all possible (infinite) environments. Brute force AI requires more development effort, but is capable of also identifying objects in many environments. This is why machine learning is a step backwards in technology and why it should never be used in life critical applications.

@Adam H Amen, I never thought of the beast as an AI! The beast will be cast into the lake of fire so I believe he will be flesh and blood human with a soul, but the 'image of the beast'!

@Adam H yes, that is a very interesting way of looking at it! I think we're a very long way from an AI being able to reason, but we've been using AI to kill people for decades.

Terrible name, though lol. Lightelligence. I get the idea behind it but it just doesn’t work as a satisfying portmanteau

Most nuclear power plants are pneumatic computers. Well, the old subs and breeder reactors anyway.

Can't wait to watch this with my kids. I forgot it was coming, we were waiting so long for it.....

NO! NO! NO! Many people say I am sick in the head. NOOOO!!!! I don't believe them. But there are so many people commenting this stuff on my videos, that I have 1% doubt. So I have to ask you right now: Do you think I am sick in the head? Thanks for helping, my dear nico

@@AxxLAfriku 0

@@zaksmith1035 That's awesome mate

@@AxxLAfriku I'm honoured to have your bot-like reply in my comment. As for the answer, well, I don't know, but have a good day mate!

Anyone knows some Good Science-Channel for me to cxheck out?

thanks a lot Brian Boni for your valuable input (keys: computers: mix of analog and digital)

@@nenmaster5218 Nile red and Nile blue

@@yuro5833 Thx! Know Hbomberguy?

@@nenmaster5218 I actually thought I didn’t then realized I had seen several of his videos and forgot about him so thank you as well

I wonder how temperature sensitive it is.

@@dorusie5 I'm mostly curious about the write-cycles and lifespan of the flash cells. Is the network going to get Alzheimer's after a few days?

integrated circuits like that have always been really efficient - the downside is that they are extremely specialized... as the guy said: its not a general computation chip it can literally only do matrix multiplication - but that it can do really damn efficient (though slightly imprecise - which likely still is good enough for neural network purposes since they arent interested in the exact value of the result) so...sure thats competitive for that one purpose - but useless for anything that isnt that purpose but if the type of calculation that they can do is in high demand - they likely can sell a lot of specialized hardware either for specific devices or plug-in cards for computers that supply fast matrix multiplication operations

I've been wondering about analogue usage of SSDs for a long time. It's an oversimplification, but each cell holds a voltage which can also be interpreted as an analog signal. If we take music as an example, you could basically write the value of one sample point to a cell, writing 16 bits worth of information to 1 NAND cell. This of course makes it impossible to compress the music, but it would allow to store music 'losslessly' at the same cell usage as a compressed 256kb/s file on TLC storage. Of course, NAND reproduction isn't perfect (and as such, music reproduction wouldn't actually be lossless), but I wonder how close this would come compared to the compressed digital file. I think this could be potentially useful for offline caches and downloads from Spotify for example, as the data can be checked and corrected when a high speed network connection is actually available.

Already? We did this before 1960.

And now, we're "evolution" but with awareness and intent heh

Critically wrong. Not one-neuron - that doesn't even make sense. One *layer*.

​@@slatervictoroff3268 I have to disagree. Perceptron is one-neuron (one neuron that receives multiple inputs and puts out one output). This makes it also one layer network I would say

@@brunsky277 thinking about it though, the inputs all had their own weights, those weights correspond to a neuron. A modern ai model has inputs, and the weights exist on the layers. Therefore the perception had 400 inputs, 400 weights, and 1 output signal. That implies to me 400 neurons, in a single layer, leading to a single output value.

@@slatervictoroff3268 No, one layer can be multiple perceptrons, it’s technically one-neuron (which is technically a one layer though)

Amazing. Just found two articles about it in the Internet Archive.

That is awesome

That's amazing. We should show him this video and ask him what does he think about it.

cap

I would love to know him

I'd love to read about that! I've been dipping my toes in ML algorithms and many of the really interesting networks require an immense amount of memory to function, on the order of tens of gigabytes. I'm curious why those models require such an immense amount of memory, and what can be done to improve that.

@@Zeuskabob1 u don’t really need 10s of gigabytes to get a good model that can perform well on a task (usually). Most people still use models of size less than 5gb or so.

+

thanks for pointing this out 🙄 seems like it was deliberately ignored for the sake of promoting this dumb startup

just another case of Veritasium making a bait video to make experts respond

Yeah, my computer classes in engineering school had a similar thing, though with us it was opamps. It was not a full class, but we did it around the same time as learning FPGAs and having to implement complex programmable digital logic, so it was a good reminder of 'digital logic with an ADC/DAC pair is not always the best or simplest solution'

While it's absolutely not the same thing, I encourage newish programmers to write a 6502 emulator. It's about as close as one can realistically get to building your own CPU hands on, which IMHO gives a worthwhile different perspective to the field than the now common approach to never leave the comfort of interpreters and virtual machines.

Hopefully he covers this topic in the future

how are u so smart

Probably because it is also an emerging system. But also because photons are used like electrons as the actor, a new actor, while the video is explaining two fundamentally different ways to act.

I was waiting for him to get to photonics, too. It's a HUGE opportunity to crazy amounts of parallel processing. And then there's quantum computing white whale, too...

I want my positronic brain patch

True, but the actual impact of this is not what you think.

Jesus, could you astroturf a bit harder please?

Until the 90s, US war ships used mechanical calculators to calculate aiming the guns, something that would be perfect for your channel.

I see what you did there.

BS! Fourier ... ROTFL

I assume there's a lot to be discovered on the topic of self-correcting algorithms, or even error-correcting analog circuits that compensate partially for the inaccuracies. Like what Hamming codes are for digitally transmitted data.

nature exists in chaos, technology is more and more approaching the chaos orchestra.

Distortion really? I am under the impression that the biggest problem with analog computer is NOISE. You can never get rid of noise in an electrical system. Even if the hardware has no distortion, the inherent thermal noise in the system will cause some small calculation error.

@@StevenSiew2 that's true, but noise is something that AI needs to deal with anyway because the inputs will always be noisy to begin with. It can actually be useful to _add artificial noise_ while training a digital NN, to avoid overfitting issues. (Stochastic gradient descent can also be seen as a way of making the training “noisy”). As long as the pertubations are small and random, training won't be affected negatively. Distortions however are hard to deal with. You may be able to train a model on a particular chip that has such and such distortion; because the distortion properties don't fluctuate and constant-but-unknown biases, the weights will ruthlessly overfit to this particular chip, and then it probably won't work at all on another copy.

As a PhD student in this field, I can answer some of your questions. Yes, we usually talk more about noise than distortion. And thermal noise is not the only issue, there is read and write variability, resistance drift in time, the resistance of interconnections, ... However, it is true that neural networks can sometimes take advantage of the noise to avoid overfitting, but only a reasonable amount of noise and only in some cases. Self-correcting algorithms and error-correcting are options, but it's not that easy. Usually, this kind of method sacrifice the performance or requires more energy (which is the opposite of what we want). About the mixing digital and analog, they presented it nicely in the video, but the digital/analog converters required a lot of energy (sometimes more than the vector-matrices multiplication itself). So we don't want to do it too often.

It sounds like your dad's computer was before the invention of the transistor. There was an analog computer at the electronics lab at the university of Hull, UK (when I was a student there in the 80s) that had moving parts. I remember when it became unstable and the professor sprinted across the lab to shut it down before it self-destructed. Something spinning suggests a sine wave generator for example.

Yeah, just like analog synthesizers. You have to let them warm up to a stable temperature first or they would constantly drift out of tune while playing. This was later solved with digital controllers.

Maybe the older mechanical calculators. Linus Tech Tips did a video on those. Super interesting stuff.

I can't address your question directly but in the later half of the 1960's I worked on a helicopter simulator, used to train military pilots, in which all computations simulating flight were performed by analog circuits made up of transistorized (no IC's) operational amplifiers and servo motors with feedback. This whole machine was housed in a 40 foot long semi trailer. In the rear of the trailer was a cockpit from a CH-46 helicopter including all the controls and instruments but the windows were frosted over so you were always flying IFR in a fog, i.e. no visuals. Next as you moved forward was an operator's station where you could control parameters such as air pressure and temperature, activate failures such as engine fire or hydraulic failure and such. The remainder of the trailer contained a row of electronics racks on each side housing the amplifiers, servos and other circuits that performed all the calculations. We can look at main rotor speed as an example of how it worked. Rotor speed was represented by the position of a servo motor from 0 to 120 degrees. The position of the motor was determined by the output of an amplifier whose inputs were derived from many variables such as engine power (there were two), collective control position and altitude. Attached to the servo motor was a potentiometer whose output drove a cockpit instrument but was also fed back to amplifiers/servos which were used to calculate engine power and such. There were many such subsystems with feedback loops interconnecting them so that failures were very difficult to diagnose. Often the only way to resolve a problem was to take a guess at which part might have failed and replace it. Also routine maintenance was very labor intensive as the many potentiometers would wear and need to be cleaned and then realigned which might take an hour for each one. As a young man I was totally amazed and fascinated by this technology. As an old man I can't believe that it really worked at all. But it did, at least some of the time.

Back in the day, circa 1957, I was an Electrical Engineering student at the City College of New York. In one of the labs we constructed an Analog Computer using physical components like Motors, Gears, etc. There was absolutely nothing binary/digital involved except weather you passed or failed the course. A couple of years later I worked with a Bendix G15 computer with an optional DDA (Digital Differential Analyzer). The DDA was an analog computer Input and output were analog. You can look upon Google. Search for " Bendix G15 computer with dda "

Derek actually made a video a few years ago explaining why veritasium would start using clickbait titles (to appease the youtube algorithm)

It got you to click didn't it?

It’s for the click bro. And for good reason. If any channel deserves to clickbait, it’s this one.

@@dinglesworld agreed 100%

its not really niche when ai and algorithms are used everywhere.

Wow nice video

Nice video

Same thought. I used this technique to figure out a way to solve mazes super efficiently with flowing water. I think that’s what’s happening with quantum computers also.

It always comes back to the drawbacks Derek mentions at the beginning of the video: Analog processing is single-purpose, error-prone, and hard to repeat. As such, for your physics example, it would invalidate A LOT of the data you get back, since you cannot guarantee a certain level of falseability, auditability, and error margins. You CAN get there, but you start requiring A LOT of boilerplate circuitry around the actual solution solving hardware. As a silly and basic example that is almost trivial nowadays, but still there, you can think of the necessity of adding a lot of surge protection and current stabilization to a circuit to ensure that the natural unsteadiness of current in the power grid won't skew your results. And that's even just taking into account discrete and "simple" issues to calculate. Imagine processing data for some chaos-related physics theory, and basically getting pure rubish at the end, because even the slightest micro-volt level of disturbance automatically distorts everything. How about external interference? Or electromagnetic interference between the actual wires in the circuit? As I said, not imposible to tackle, but you suddenly have an overhead of 90% boilerplate just to make the results useful on anything practical. I can't even imagine all the engineering that must have gone into those Mythic semi-analog chips for AI, just to keep everything tidy. The fact a Realtek-sized chip can give you one third the performance of some nVidia Quadra (or similar) card, for a fourth of the power consumption of a cheap entry-level mobile Core i3 is just astounding!

To be clear, these Mythic chips point towards a future resurgence of analog processors not dissimilar to what digital ones brought in with their unparalleled versatility. Outside of very bespoke chips for very high amounts of money, probably in the realms of very high research, science, etc; I can see a general idea of modularity at a functional level. Say, you manufatcure analog chips that can do some very important but expensive math calculations that are common for most science in some specific branch (say, a lot of transformations, or integration, maybe some Lorentzians, and so on). Then, at research groups, institutes, university, they go and do the same as electronic engineers do with good old breadboards, and DIY some complex formulae on the fly, test their hypothesis, and iterate over the formuale as needed. Imagine those astrophysicists doing 2k term polynomials, being able to duct tape a dozen chips together, the same way electronic engineers use logic gates as basic digital units, and getting the results out in a couple of hours, instead of having to write a piece of software that will take a couple of days to run, a week to write, and any mistake or failed result requires another week to debug just to make sure it failed because you were wrong, and not because you input a 5 where a six should have gone when writing all 1500 terms for one of the formulae

Yes I've always thought this too. Aside from the negatives mentioned in the vid and comment above: "if you can't calculate it, let nature do it" I've used the term 'calculate' here, but I think it applies in a broader sense. If something is too hard to manufacture / produce precisely maybe nature can do it better.

@@yoshienverde wish I understood what you were saying but great rebuttal

most of his videos are like this lol

I believe Derek found a little niche to focus on since he did the video on the ancient Greek analogue computer, which had an almost identical conclusion

I feel he is on to something here ... maybe the real benefit is that you don't have to make all these specific chips, because in principle one fairly big analog one could do everything you threw at it. But it feels a bit scary to me too, because you are getting closer to biological systems.

Yeah, the reduction in power consumption I'd imagine is mostly due to it being an ASIC, not being analog. There are quite a few digital AI inference ASICs coming onto the market as well - I'm curious to see which ones will reign supreme

Hi: Unfortunately I can not speak English, so I have to use the program I have installed on my phone for translation, except that I can speak Persian, I do not speak any language other than Persian and no other language. I can not speak, so if I said words or sentences and the special program mistranslated into your language, I apologize in advance to your esteemed father, from the bottom of my heart and from the bottom of my heart, that the letters and words Call me to the end and call me a person orDo not consider me a rude person and do not consider me a rude person and judge me correctly and after reading my writings, just put yourself in my place for a few minutes and imagine yourself in my place, maybe if you have an awakened conscience and There was love and affection in your hearts, of course, if you did not have pride and arrogance, understand me and give me the right, and again, maybe, maybe, maybe you did something and you did it for me and you took me from You saved this great tribulation that I hope will not happen to any living thing, of courseI do not very much hope that anyone will take my hand and save me from this misfortune, but I am writing so that I can at least be comfortable in front of my conscience and not blame myself later if I cry out for this cruelty, oppression, and captivity that has fallen on me. All the doors are closed to me, even the eyes of God are blind to see the oppression that is being done to me, and the ears of God are blind to hearing the cries of my constant cries, my midnight cries, my daily sufferings, my daily prayers and the jurists and The cries of my every moment from this oppression, oppression and cruelty that from the beginning of my life, fromThe first events of my life that I remember are deaf. So how can I hope for others when God has done nothing for me and trapped me in a cruel, cruel, cruel, and oppressive family? I am Morteza, I am from Iran, I am blind, I am 34 years old, I am unemployed because in our country, Iran, there is no work for healthy people, let alone disabled people like me. I live with my family in a small town in Iran. Of course, in appearance, they are my family, but in reality, they see me as their own enemy, and with me, who am their child, only bI was born blind, and I was not to blame for this, but my parents, because they were illiterate, considered me as a disgrace only to myself for my blindness and disability, and were always tortured as severely as possible. They beat me to the point where they threatened me with a knife. They put me to sleep and put the knife on my neck, and I was terrified and scared. Get up and until fullThe thin skin of my hand burned, they would not let me go. And they tortured me many, many times, to the point that my brothers, with the support of my parents, tortured me in front of my parents' indifferent eyes, and told me that I was blind. You are and you should be tortured to the extent that they created the belief in me and in my mind that anyone who is blind or disabled should be tortured because it is a disgrace to the family and society. And unfortunately, their torture is still going on, and only a kind of tortureThey do not consider it for him, so how can I complain to someone who considers my law as his property, of course, the current laws in Iran. Unfortunately, the government does not have a place for people like me to go and live. I really have no choice, either I have to commit suicide or I have to stay in the same house that my family has turned into hell for me, under the severe mental and physical torture that my family inflicts on me. And more than ever with a horrible gradual death that tormentedI want to stay. So if you still have a little mercy, fairness, conscience, compassion, love, humanity, knowledge, ideology, humanity and altruism, help me, hold my hand, reach my cry. If you are in contact with institutions and organizations affiliated with the International Committee of the Red Cross, or if any of you are a member of human rights organizations or the International Committee of the Red Cross or Red Cross organizations in free and pro-human rights countries, listen to my writings, my voice and my cry. Come on, maybe they're a little fair and think of me and a way toSave me from the clutches of ostensibly wolves, of course, if human rights institutions and organizations and the Red Cross actually support human rights, if you are a citizen of a free country or a citizen of a free country like the United States, Canada, Australia, Great Britain, Norway, You have Sweden, Denmark, Finland, and the European Union. Send me invitations, arrange for me to leave this house, which is worse than hell, so that maybe I too can taste freedom. If you have capital, you are rich, you have money, at least help meLeave the messengers, so that all the photos on my personal pages in the messengers and social networks are from 8 years ago, that is, for 8 years ago and 5 years ago. I have not even traveled for many years, because our city is a small city. It is sad for a disabled person to leave home, it is sad for a disabled person, the streets of our city are not adapted so that a disabled person can easily be at least a little out of home. And I'm really depressed at home, especially with this family that is always torturingAre. Of course, I do not care if I travel or not, because my problems are so great that not traveling travels much more than my other or other problems. In your opinion, can a person who eats only one meal in 48 hours, which is a good meal, even a moderate meal, but also any garbage he gets, take pictures, travel or not, and other things? Slowly, someone who breaks his heart at every moment, of course, the work of my heart is no longer broken and my heart is on fire andIt is burned, and this fire is getting more and more hot and burning, and it is burning and ashes my whole being. So I hope that if you do not hear the sound of my heart breaking, at least be fair and feel the smell of my heart burning, away from arrogance, arrogance, misguidedness, ethnicity, race, skin color, shape and appearance for the sake of humanity and for Humanity and honor that you have, just imagine yourself in my place for a few minutes, see yourself in my place and think, if far away from you, God forbid, you would be in my place.Did you ??? Were you satisfied that someone was making fun of you or disrespecting you, or did you laugh at you in response to your message and pain and heart ?, so my dear friends and those who wrote my writings to you Please, if you do not want to do anything for me, at least do not make fun of me, do not laugh, and if you do not want to help extinguish the fire inside me that burns my heart, at least do not spill oil and gasoline with ill-considered language, words and expressions, and this is far from knowledge, honor, It is humanity and family originality. If you want to give me anyPlease help me do not send me a message on RU-vid because I can not transfer messages from RU-vid to the translation program that I have installed on my phone, and as a result, because I am not fluent in any language other than Persian, I can not understand the meaning of messages. Let me know what you sent me on RU-vid. If you wanted and could and the possibility of helping me in any way, whether financially, materially or spiritually, please contact me on WhatsApp or Telegram, because in WhatsApp and Telegram I can easily send messages.00989358205228 00989182804420 00989358205228 00989182804420 00989358205228 00989182804420 00989358205228 😞😞😞

Fancy seeing you here ❤️

hm

hook, line and thinker

Damn Canadians keep blowing my mind. TELL THEM TO STOP IT!

SARS CoV-2 was patented with UNITED STATES after being developed by Pirbright Institute in UK

I was waiting for him to either mention the words "neuromorphic" or "memristors"

I remember reading about Mead's analog stuff in the 1980s, something related to hearing. Perhaps my memory is wrong.

Exactly! The analog behavior of neurons is closely modeled by the analog current/voltage exhibited by the tweaked transistor cells, as so well demonstrated and visualized in the video.

Neurons have a lot going on inside, and things are happening outside, that affect what they do and when they do it. It amazes me that computer neural networks work at all, let alone as well as they do.

Ya, so take what he's wrong about in the field you know and apply it to the field I know(comp sci) and suddenly the entire video is a load of rubbish.

@@kalliste23 Don't get me started about the neurons in a squid vs a human... they have fewer, bigger and more complex neurons.

@@vyor8837 Not a load of rubbish, just amazingly primitive compared to what it's trying to mimic. And also use specific. And not really capable of learning new kinds of tasks. ;-)

This is exactly the point though. Trivial for a human to do, hard for a computer.

@@henrypetchfood I literally just had a friend tell me a story about their mother misidentifying a pomeranian as a cat haha. Maybe not always trivial.

@@henrypetchfood It's trivial for a human because evolution produced neural circuits capable of solving this very hard problem. Our own minds are the least aware of what they do best.

True, but Derek's point is that in the grand scale of what we'd hope to achieve with analogue computers (in the future), telling apart cats and dogs is a simple expectation - yet it's still hard to do with current technology.

quantum computers (one that use physical q-bits) are only hypothetical, but people talk as if they already exist in reality. They don't, there is not a single, fully functional quantum computer on the planet, and there might never be.

@@BreaksFast Sure they exist, they just don't have very many q-bits.

@@BreaksFast The computers do exist, but there is a lack of understanding in programming them to do classical computing problems.

@@BreaksFast They can and do exist, albeit with very limited qubit counts. The first experimental demonstration of one was in 1998. D-Wave Systems are selling computers with 2048+ qubits right now.

@@BreaksFast Tell that to IBM?

Speaking of radio receivers, think SDR. software defined radio, a revolution in radio..

Quantum computers are not right around the corner

Quantum computers are not gonna be beside your computers unless you have a massive shed with space for a state of the art helium/nitrogen cooling system on hand

@@Kyrator88 I'm not completely discounting the possibility of solid state room temperature qubits, but yeah, excitement about personal quantum computers is pretty silly.

A quantum computer is not required for performing quantum-safe encryption/decryption. NIST is very close to standardizing one of many candidates that provides this functionality.

Oh my god thank you for bringing this up. Ever since i watched that show I have been set on learning everything and anything about AI. It has inspired me set me on my current course in Comp Science.

I gave you your first like 😁

@@Mani_Umakant23 Why would you like such an unoriginal comment that provides so little value or thought?

@@N____er Aise hi sexy lag rha tha.

@@N____er Don't say anything bad about ElectroBOOM he is such a wonderful creator

Hi sir I am a big fan of yours \

That's pretty creative and clever of him.

opamps are wonderful

That sounds like the micro-mouse project. In the UK at schools in the late 90's it was a challenge for electronics students (electives yr10-12). Really fun finding solutions to the problem of navigating a maze either on paper (a line) in a 3d maze or both.

Your comment was stolen :/

Intersting

@africancheez Engineering is a great profession. Along with Analog Computers we had to take classes with Digital Computers. As a student, I used the HP1100 (CPU , paper tape reader, 1MB hard disk), IBM370 (teletype and punch cards), and PDP11 (punch cards) to learn Fortran IV, Pascal, and Basic. My professors were experimenting with the TRS80, Timex, and Apple 1 personal computers (most arrived as kits) as a hobby (top speed was 1MHz...1977 timeframe). In the computer lab, there was a desk top PC made by Commodore (commodore PET with a built-in cassette tape for mass data storage), but only Grad Students were allowed to use it. It was good to know the difference between solving certain problems using a program on a digital computer versus solving the same problem using a far simpler analog computer. good luck to ya.

I can only speak for my associates degree program but we still do that. This past semester we used IC opamps to make integrator and differentiator circuits. They were some fun labs. I love seeing some of the signal art people make on oscopes.

@africancheez "we have had the technology for a long time, but the world just wasn't ready for it yet." It was never just a technology problem. Do you want your data in someone else's warehouse? Very often, no. Actually I believe it is never optimum since you lose some 4th amendment rights regarding search and seizure. On the other hand, a small company that cannot justify the expensive and reliable talent needed to guard company secrets has no good choices and might as well go "in the cloud" but even there dividing up the cloud (some to Amazon, some to Microsoft) would prevent or reduce complete and total disaster.