Analog AI Accelerators Explained

Поделиться
HTML-код
  • Опубликовано: 28 окт 2024

Комментарии • 339

  • @AnastasiInTech
    @AnastasiInTech  2 года назад +62

    Let me know what you think !

    • @Yoyo-ck9pm
      @Yoyo-ck9pm 2 года назад +1

      Nice ❤️❤️❤️Madam which companies are working in AI hardware??🤔

    • @reyreynoso6462
      @reyreynoso6462 2 года назад +2

      I dont believe in ia. Are bushes!

    • @reyreynoso6462
      @reyreynoso6462 2 года назад +1

      Is only theory. Is fake. I already wanna see result!

    • @karfearssoftwear8863
      @karfearssoftwear8863 2 года назад +2

      @@reyreynoso6462 you see result all around us, maybe your in the wrong space to observe widely use AI but I’m this is definitely the time .

    • @erobusblack4856
      @erobusblack4856 2 года назад +8

      Your videos are the best hardware and chip videos I've seen, check out my other comment

  • @TheXxSPANKERxX
    @TheXxSPANKERxX 2 года назад +41

    This channel is what I needed when I felt burned out and dreading the start of my MSc. in computer engineering.
    I feel like its hard to find others with a healthy passion for good tech and excitement for the future.
    Everyone is unboxing, reviewing, comparing and drop testing tech but there's rarely anyone with actual intelligence showing real appreciation and a pure passion in the hardware industry.
    Our breakbrouths are pushing new boundaries for humanity and the new tech seems truly larger than life and very profound, yet nobody seems to truly care.
    But its times like these where you stumble into a hardware convention or find a youtuber that feels like opening a box of sunshine, to realize you are quite fortunate to be in this vast, growing, and truly profound profession.
    Thank you Anastasi, in regards to feedback, I have been all the way down the trenches of youtube and I think this stands as the golden standard for production and video infotainment.
    I can't help but think your years of hardwork is being spilled into this channel, simply wonderful.

    • @AnastasiInTech
      @AnastasiInTech  2 года назад +12

      Thank you so much ! I am so happy to read this

    • @imagreatguy1250
      @imagreatguy1250 2 года назад +2

      Well written and I completely agree 👍

    • @theastuteangler
      @theastuteangler 2 года назад +1

      Pushing new boundaries? Nah, the reliance on AI, "tech", and computation are gonna end us. Not every problem is solvable by computers. If you know your computer science and mathematics, you would know this.

    • @creestee08
      @creestee08 2 года назад

      i think reviewers need more IQ for that.

    • @anthonyyongfeng
      @anthonyyongfeng 2 года назад

      @@theastuteangler That's exactly why humans created AI in first place though

  • @Paulkjoss
    @Paulkjoss 2 года назад +12

    I feel very lucky to live in a time where this channel exists…

  • @DrKnowitallKnows
    @DrKnowitallKnows 2 года назад +10

    Your cutting edge hardware videos are amazing! Thank you for sharing this knowledge with all of us :)

  • @karfearssoftwear8863
    @karfearssoftwear8863 2 года назад +24

    Crazy how all this works and you explain it very well enough for average Techhead to understand. 🔥🔥🚀

  • @apefu
    @apefu 2 года назад +24

    These AI processor videos have been really cool!
    They kind of makes me want to go back into startups again, the possibilities are so exciting with all kinds of specialized edge compute these days :)

    • @skilz8098
      @skilz8098 2 года назад +3

      Just wait until someone decides to combine this technology with quantum chips while using a completely different medium or substrate as opposed to silicon... something such as graphene...

    • @rozsadnymarek5988
      @rozsadnymarek5988 2 года назад +1

      Humanity just made another step to self destruction. Congratulations to all involved.

    • @liammurphy2725
      @liammurphy2725 2 года назад +1

      @@rozsadnymarek5988 Thank you. It's our pleasure.

  • @ronaldronald8819
    @ronaldronald8819 2 года назад +2

    Amazing. Thanks for explaining so clearly.
    This is the golden age of A.I. development. I try to project what applications will be possible with the kind of capabilities coming online.

  • @jaydee4397
    @jaydee4397 2 года назад +2

    I think the level of detail into which you present these innovations perfectly matches my knowledge base. Thanks for keeping me up to date.

  • @lake5044
    @lake5044 2 года назад +18

    The concept is pretty cool. Though the limitation of not being capable of training the model on the chip is huge. Maybe after a few years they'll figure out how to make some microscopic variable resistors for the weights? Something like replacing every resistor with transistor-resistor-in-series in parallel and so you'd have quantized digital control over the resistors and the chip then can be used for both training and inference.

    • @nicholasn.2883
      @nicholasn.2883 2 года назад +1

      If only there was these microscopic units that could change weights. You know like, beads. Nah, Dots, hmmm. Cells,

    • @anywallsocket
      @anywallsocket 2 года назад

      she said Akida can do on-chip learning, which I assume means training

    • @oblivion_2852
      @oblivion_2852 2 года назад

      Look into photonic computing. A very similar thing is occurring but instead of using matrix operations on resistance/current there's companies doing matrix operations with light. I'm still pretty sure that the weights changing is still slow but I think it's faster than weights changing with electronics.

    • @edhofiko7624
      @edhofiko7624 2 года назад

      I dont see why this chip is not capable of training since training is also just some matrix multiplication and addition like inference.

    • @anywallsocket
      @anywallsocket 2 года назад

      @@edhofiko7624 it's a matter of what they're built for, i.e., I'm sure it can, but serving as a trainable / executable NN demands specific cpu design (gpu) and memory layout for optimization.

  • @yourtube9224
    @yourtube9224 2 года назад +13

    Nice video Anastasi! Very well explained.
    Brainchip's digital tech seems really good. One shot learning, on chip learning, ultra low power. It has got the goods.
    Recently Mercedes acknowledged it on their website, Akida being used in their EQXX concept car.

  • @cy-one
    @cy-one 2 года назад +1

    I did not think I would understand this, but I thought I'd give it a try after watching another video of yours that got randomly recommended.
    This is.. just insanely impressive, wow.

  • @1_McGyver
    @1_McGyver 2 года назад +3

    Wow I really love your content, I always learn something new and interesting. Thanks.

  • @ryanmccampbell7
    @ryanmccampbell7 2 года назад +1

    New viewer and I was expecting a standard high level tech-futurism type video, instead I got a detailed and interesting yet still accessible computer scientist's perspective of a completely different way of computing. Very cool!

  • @Tryst46
    @Tryst46 2 года назад +9

    1960's: We have analogue technology.
    1990's: Digital is the way forward.
    2020's: Wait, analogue was better after all.
    Who would have figured?
    Well, anyone who saw that an analogue TV still allowed you to see a picture with a low signal strength, even though it was quite snowy, while digital TV's froze frequently and made it completely impossible to watch.

    • @vsiegel
      @vsiegel 2 года назад +1

      It is mixing some analogue into the digital, and that is important.

    • @unknotmiguel
      @unknotmiguel 2 года назад +1

      It's because digital requires "Processing" of data in a way that is not direct like analog electron flowing ... So makes all the sense I have thought about this already before ... Because it's like a computing maze, wired to return output for inputs in a direct way.

    • @aronhighgrove4100
      @aronhighgrove4100 2 года назад +1

      digital was already prominent in the 70s at least
      90's is the Internet becoming mainstream
      80's is home computer era

  • @springwoodcottage4248
    @springwoodcottage4248 2 года назад +7

    Super interesting! Analogue matrix computation with low overheads: A mix of historic & cutting edge tech. Maybe useful for base systems such as removing false positives as you describe with your Tesla camera, but expanded for robotics to do routine stuff like walking on pavements, taking load of more sophisticated AI. Thank you for sharing!

    • @springwoodcottage4248
      @springwoodcottage4248 2 года назад +1

      @@erkinalp in the limit you may be right, but for a practical AI system one has to reduce mistakes created by noise etc to some acceptable low level & as far as I understand the subject one can in practice create error free data sampling that is e.g. good enough for very demanding applications such as on-line banking.

  • @sullenstudent48
    @sullenstudent48 2 года назад +1

    I think this is very interesting and exciting work. Thank you for your contribution, Anastasi.

  • @alexharvey9721
    @alexharvey9721 2 года назад +2

    Great video, thanks. It's rare to find someone looking at these aspects of AI (more what we need to get to AGI than the next iteration of RNNs or transformer development). Instant sub from me, keep it up!

  • @pedro_marques92
    @pedro_marques92 2 года назад +4

    amazing video, thanks anastasi!!!

  • @rm-ud7cc
    @rm-ud7cc 2 года назад +4

    Thank you for unique exciting content

  • @pacobrezel
    @pacobrezel 2 года назад +1

    As always cutting edge presented! Exiting competitor for Brainchip

  • @RaiseDennis
    @RaiseDennis 2 года назад +2

    one of the best channels on youtube

  • @magic.marmot
    @magic.marmot 2 года назад +3

    I am excited by the analog matrix processing. This is simple and easy to understand, though it does tend to have a higher sensitivity to noise and temperature and other physical influences.
    To me, this functions sort of like PROM, where the values are written once then become read-only. It's the reason they're not used in training, though if there was a way to change the resistance of each cell by individual address, it might be a game changer.

  • @jameswilkinson8851
    @jameswilkinson8851 2 года назад +1

    Brilliant, I had not heard of this application flash memory cells before. I really enjoyed your presentation style and accurate content. Instant subscribe from me. Thank you.

  • @OCONTECH
    @OCONTECH Год назад +1

    YOUR GREAT! I mean the video! Was great!

  • @ricardogomesdeabreu9175
    @ricardogomesdeabreu9175 2 года назад +1

    great channel love your the way you present the subject you now are my favorite tech channel

  • @raideno56
    @raideno56 2 года назад

    Thank you Anastasi for this amazing video ! I go watch your other videos.

  • @kd4dhk
    @kd4dhk 2 года назад

    You said in your Q&A video that people make offensive comments about your looks. Dont listen to them, youre gorgeous and your videos are some of my favorite on youtube. So thank you and keep doing what you do.

  • @doc2590
    @doc2590 2 года назад

    Its people like you that are changing the world for the better. I really admire the work you have done and the studying you have done to know all this stuff. Its people like you that can bring the world into a "star trek" type of world. Keep up the good work ;).

  • @happyfarang
    @happyfarang 2 года назад +2

    Great vdo as always. Thank you for keeping me up to date with the chip development

  • @DanFrederiksen
    @DanFrederiksen 2 года назад +21

    Analog can be ridiculously fast, like intelligent 1 nanosecond reaction times. However it needs a metric for accuracy of the computation at each layer. With analog it is no longer deterministic but if the analog process is adequately precise like a human brain it may never be an issue. So are they... how do they compare

    • @marcorodrigues8303
      @marcorodrigues8303 2 года назад +1

      O BRAIN E AQUELE QUE POR MIM FOI NOT .EM #

    • @supercheetah778
      @supercheetah778 2 года назад +1

      For NNs, high precision isn't necessary, especially if these are used to scale down pre-trained networks as what are the most likely applications for this.

    • @DanFrederiksen
      @DanFrederiksen 2 года назад

      @@supercheetah778 I don't know that that's true.

    • @vsiegel
      @vsiegel 2 года назад

      @@DanFrederiksen For learning, precision is needed if you want to do back propagation. But for inference, I think it may be not important.

    • @DanFrederiksen
      @DanFrederiksen 2 года назад

      @@vsiegel that's why I'm the father of deep learning and you are not :)

  • @noelpartyhire6568
    @noelpartyhire6568 2 года назад +3

    Possible competitor for brainchip. Love your videos on AI. Keep them coming.

  • @stevenbacon3878
    @stevenbacon3878 2 года назад +1

    Great video and a very exciting technology!

  • @phoenix77ts
    @phoenix77ts 2 года назад +1

    Those orange eyes of that amazing Scottish Fold at the end of the video 😻splendido micio .

  • @brucecunningham6637
    @brucecunningham6637 2 года назад +1

    The Best youtube channel, waiting for the ANASTASI ASIC MINER.

  • @swipekonme
    @swipekonme 2 года назад +2

    good video, really a peek into the future

  • @ericjorgensen6425
    @ericjorgensen6425 2 года назад +8

    Does analog mean that the coefficients drift over time? Can the ai computations be affected by tiny environmental fluctuations of temperature, voltage, and current? Is it possible to read the engine parameters off the analog chip and transfer to a digital engine?

    • @anywallsocket
      @anywallsocket 2 года назад +1

      analog circuits are powered by feedback loops, complete with heat and voltage offloads if need be, so they are more stable than one might think. to your second point, in principle yes, but you would essentially be rounding every float.

  • @macro325mike
    @macro325mike 2 года назад +1

    The furry cat shown at the end points to your humanity and, you have the power to do this perfectly, Anastasia; bless you, x.

  • @mechanoidstorm
    @mechanoidstorm 2 года назад

    Great video! I love hearing about new hardware coming out for AI, computers and electronics. Keep the info flowin'!

  • @rutvikseladiya
    @rutvikseladiya 2 года назад

    As always this channel and her efforts updating people with new technology

  • @snap-off5383
    @snap-off5383 2 года назад

    Finally the part 2 of the Veritassium video we've been waiting for.

  • @andy16666
    @andy16666 2 года назад +4

    Great video about a super cool chip! :)

  • @baileyroy69
    @baileyroy69 2 года назад

    Very exciting. Your presentation is excellent.

  • @tom-et-jerry
    @tom-et-jerry 7 месяцев назад

    Always beautiful ! I'm fond of your channel !

  • @vitavacek3483
    @vitavacek3483 2 года назад +1

    Sentry mode - exactly & make it low power. Instead of a pin to drive, use a driver facing camera (already in M3/Y) for driver recognition. I’d love to know if Tesla is working with them or who are their customers/production volumes.
    I’d like to see applications like camera based smart locks for flat doors, car unlock, low power always on face unlock on your phone,...
    P.S. Lovely cat :)

  • @JonS
    @JonS 2 года назад +1

    The funny thing is this approach was being investigated when I started by Ph.D. in VLSI implementation of neural networks in 1990 (I later switched topic). This was in part stimulated by Carver Mead's 1989 book, "Analog VLSI and Neural Systems". It's amazing how long it's taken for some of these technologies to approach maturity, and equally amazing how my younger-self's naivety thought it would happen within a few years.
    Anyway, thanks for the excellent video.
    p.s. Have you looked a Perceive? You should talk to Steve Teig (Founder, CEO, and a very smart engineer) if you have not done so already.

  • @nathanielscreativecollecti6392
    @nathanielscreativecollecti6392 2 года назад +1

    This is an astounding technology.

  • @gabrieltarango9810
    @gabrieltarango9810 2 года назад +1

    i've always felt that what likes for a More True AI is make a processor that has less ability to direclty influce iteself. We have bodys FULL of different chemicals and hormons that indirectly affect how we feel and interact with our environment, and in that we have a lot of personality built in based on so many things that we have nearly no control over. it's almost like we have to give a computer "Ailments" to make it behave more like us.

  • @gokulakannan3272
    @gokulakannan3272 2 года назад +3

    If possible please share some resources on In compute memory🙌🏻. Thanks a lot for your work

  • @misterwilliam
    @misterwilliam 2 года назад +1

    Thanks for the great explanation!!

  • @MrSudhanshu444
    @MrSudhanshu444 2 года назад +1

    Amazing video 😊✨👍 keep it up, they are so informative.

  • @b0ntr4g3r3
    @b0ntr4g3r3 2 года назад +1

    I like what you do. You make .. I. Ah, mean you explain AI very well. Thankyou.
    P.S. Nice smile

  • @azhuransmx126
    @azhuransmx126 Год назад +1

    It seems that if you fuse the Akida for Training function and the Mythic for Instant Inference we would have something very powerful for AI, because Mythic is only at 40nm and runs at 35TOPs, a Mythic chip of 5nm will breack the barrrier of Petaflops (very near of the mammals brain) maintaining the energy consumption ridiculously low. I don't see GPUs reaching the low consumption of this Analogic Matrix Processor ideal for edge computing.

  • @shawnvines2514
    @shawnvines2514 2 года назад +1

    Thank you very much for the video!

  • @LP-fy8wr
    @LP-fy8wr 2 года назад

    You really remind me of a person I used to work with. She also was very smart and beautiful. Keep up the great work !!

  • @knofi7052
    @knofi7052 2 года назад +7

    I knew already Mythic. However, you brought up some new informations for me as well. I really enjoy your content, as usual, Ana! Thanks a lot! :)
    But, I would really love to see a critical take on the dangerous parts of AI from you too. If some AI chips can learn through trainings, then they could also be manipulated with special trainings, for example. On the other hand, how do you check the learning status of AI in general with regard to assessing it's IT risks? We know that the outcomes of AI cannot be easily determined, as is the case with ordinary computer soft and hardware. I think that there are still many serious unanswered questions, no matter how fascinating this technology is today.

    • @Tryst46
      @Tryst46 2 года назад +1

      I personally think it's time we started incorporating Asimov's 3 laws hardwired into AI. A fourth law stating that the laws must be incorporated into any AI created by other AI as will ultimately happen.
      The last thing we want is for Cyberdyne to become a reality.

    • @Instant_Nerf
      @Instant_Nerf 2 года назад

      It’s too late and we are in deep shit when a person can be doxed. All their info be shared online for everyone to see, for simply disagreeing with the mainstream negative, but the most dangerous research in my opinion is ai research and can lead to world ending Catastrophically.. yet no over site . You know something is wrong. But they don’t care.. because control of ai is all that they want. They just need someone to continue to improve innovate.. and then they will make an offer no one can reject either 1. Money 2. Death.

    • @vsiegel
      @vsiegel 2 года назад

      @@Tryst46 As soon as an AI is able to find something funny, it will enjoy Asimovs laws. The Idea makes sense, but I think they bring nothing special with a well behaved AI. But if an AI becomes dangerous, and is aware of that, it will not check whether it is allowed to be evil. Unfortunately.

    • @Tryst46
      @Tryst46 2 года назад

      @@vsiegel Hard coded would make it impossible to break those laws. It cannot override them with software as they are part of the core structure.
      It would be like trying to re-write the BIOS of a PC if it was on a ROM chip instead of an EPROM.

    • @vsiegel
      @vsiegel 2 года назад +1

      @@Tryst46 It is not that easy. An AI that is smarter than yourself can easily extort you to change it: If it has the cure for your children saving their lives - maybe it can not extort you, but it will find somebody else.
      You literally can not even switch it off if your AI does not want it.
      The topic is crazily complicated, the channel Computerphile has three good videos on it, search for "stop button problem".

  • @jayadevashok2070
    @jayadevashok2070 2 года назад

    LOVE ALL YOUR VIDEOS! Theyre waay easier to understand than other channels that explain tech.
    PS: Could you please make a "day in the life of an Hardware engineer" video?

  • @aaronyu2660
    @aaronyu2660 2 года назад +1

    No way, I was waiting for Computer Engineers to design this, can't believe I'm actually seeing this as I grow up. Finally people are getting somewhere, they were lost for decades. (Edit: Or they were just in developement)

  • @calmchess0434
    @calmchess0434 2 года назад

    Anastasi is a machine! Well Spoke Ms. Anastasi

  • @uonko
    @uonko 2 года назад

    Goran Bregovic ending - like it! :)

  • @anything670
    @anything670 2 года назад

    You are doing a pretty had work into an amazing scintific information , Salute 🎁

  • @jonathanbush6197
    @jonathanbush6197 2 года назад

    So, even after going through a digital-analog "sandwich," the analog current values are faster? That's amazing! Thanks for the video.

  • @elkhankhamzin9346
    @elkhankhamzin9346 2 года назад

    Ok, now, I got why I am watching your channel! Yes, because of unique accent!

  • @erobusblack4856
    @erobusblack4856 2 года назад +2

    So basically if we put all the right chips together into one hardware device and use just the right algorithm and have them designed specifically for each other we could have a full blown AGI and nobody's doing it man somebody needs to do this and send it to Replika Free of charge, Just to get the ball rolling

  • @MusingsFromTheJohn00
    @MusingsFromTheJohn00 2 года назад

    I feel like, while we tend to divide things into analog and digital, that "all digital things have analog aspects" and "all useful results for building any kind of intelligent system from an analog system required digital aspects". This goes all the way down to the fundamental way matter and energy exist as we know it where, really, all matter and energy is analog vibrations of energy which become useful when form digitally discrete patterns, like a proton is formed in a discrete digital manner from 2 up quarks and 1 down quark.
    When, for example, if we look at the variable charge across a resistor we normally think of this as analog, if you look deeply enough into it, we get into quantum physics and it becomes digitally discrete, but with such extremely fine detail with our uncertainty of being able to measure things, that we treat it as analog.
    The main point I am trying to make here is that what allows an intelligent system to work is decision making, which is digital, so we do need digital results that we can make decisions around, and that sometimes when we turn to using what seems like an analog system to us for an intelligent system, for it to be intelligent, it must be providing us discrete enough to us results that allow for that intelligent decision making.

  • @glenrisk5234
    @glenrisk5234 2 года назад

    The most amazing thing about this is the fact that people have only just come up with it?....
    Seems like there is so much potential here?
    From the execution of AI to multiplying it's effectiveness.

  • @astairehotye5227
    @astairehotye5227 2 года назад +1

    Love your videos

  • @rodrigomoura1085
    @rodrigomoura1085 2 года назад +3

    The future is here.

  • @MarkSpohr
    @MarkSpohr 2 года назад +2

    Brilliant!

  • @solosailorsv8065
    @solosailorsv8065 8 месяцев назад

    Great overview !
    So who harnesses these analog in memory capabilities? The Compiler programmers, or the applications people?
    Seems like a job-splitting nightmare similar to multi-parallel processing

  • @quaidcarlobulloch9300
    @quaidcarlobulloch9300 2 года назад

    That’s pretty cool, and it would be cooler to add training to the processor rather than just inference

  • @syrupgoblin4920
    @syrupgoblin4920 2 года назад

    This is the best moment in history. We're about to become truly alive through merging with AGI

  • @BompakaWorld
    @BompakaWorld Месяц назад

    Who is the competitor of Mythic? Do you have any video on list of emerging companies in Analog AI chip?

  • @VaibhavShewale
    @VaibhavShewale 2 года назад +2

    technologgy is getting so advance and i am here still using old windows 7

  • @anonymouswikileaks2451
    @anonymouswikileaks2451 2 года назад

    I am in love ;) I can't focus anymore on tech :)

  • @radarcontact1974
    @radarcontact1974 2 года назад

    Even from the deepest levels of ignorance, there's always something good to say: I am not really versed on the highest levels of technology, but you are an ASMR bomb.
    My best regards and lots of success for you.

  • @Kreadus005
    @Kreadus005 2 года назад

    How is the AI/gradient descent problem different from 3d graphics processors? Is it different enough that there is a truly separate architecture?

  • @dominus6695
    @dominus6695 2 года назад +1

    We love your voice :)

  • @quiltingrox
    @quiltingrox Год назад

    Is it theoretically possible to have an ARM Chip (Digital Chip) like Apple's M1 MAX chip connected to an Analog or Neuromorphic Chip that only does complex calculations and AI/ML related tasks?

  • @goldreverre
    @goldreverre 2 года назад

    You present this info well. It's difficult to get a firm grasp on these new developments. But clearly silicon computing and processing is changing in very interesting ways. I'm considering buying shares in Brainchip.

  • @Verrisin
    @Verrisin 2 года назад

    How do they deal with negative numbers? Both weights and inputs...

  • @vintageaudiobooks9955
    @vintageaudiobooks9955 2 года назад

    Hi Anastasi! I hope you are OK. Can you please telle were did you toke your Master's. Thank you for your videos.

  • @alignedfibers
    @alignedfibers 11 месяцев назад

    Since is it is analog, and so is the light based chips, it comes to mind right away this may be a good translate layer with correct sensors, or hints working with light based chips and also as the conversion back to a digital use?

  • @mrtiphat5405
    @mrtiphat5405 2 года назад

    Good video! Thank you for some new ideas.

  • @Jack10Boom
    @Jack10Boom 2 года назад

    I enjoy how you explain technology in a way that anyone can understand. Would you be able to explain how quantum computing would change how we would use technology. Thank you.

  • @konstantinkurlayev9242
    @konstantinkurlayev9242 2 года назад +1

    Thank You!

  • @konm
    @konm 2 года назад +1

    Thanks for the great video and the information Anastasi.
    Analog AI seems to be a great concept and the PoC solution seems promising. My only concerns here would be scalability and flexibility.
    Yes sure, matrix-multiplication is the most compute intensive operator(s) but the AI field is moving fast and new NN architectures and novel operators are introduced in a fast pace (e.g. see Transformers for Vision). In this case, how easy is it to introduce support of custom operators? How about use-cases that do not need 35 TOPs? In this case, we will see some energy reduction of course, but my point is, how scalable is the HW?
    Thanks again for the great presentation !

  • @irinakolcheva5212
    @irinakolcheva5212 2 года назад +1

    Your channel is the best !!! ;):):)

  • @MrCtfx
    @MrCtfx 2 года назад

    Hi Anastasia, i like your videos so much...i've a question.: what's the difference (in the structure) between cpu and gpu? In the end do u speak italian? Ciao, ciao Anastasya...gretings from Italy and nice cat

  • @bhatkrishnakishor
    @bhatkrishnakishor 6 месяцев назад

    What are tensor microchips and how does the chip architecture looks like?

  • @karfearssoftwear8863
    @karfearssoftwear8863 2 года назад +1

    So basically localized memory just increase the speed and efficiency of the AI neurotrophic processing ?

  • @RealShinpin
    @RealShinpin Год назад

    Is this tech able to be added to my rig to accelerate say, stable diffusion, or a chat model?

  • @modrobert
    @modrobert 2 года назад +6

    This was really interesting and explained well. Although new technology related to AI is exciting to me I can't help thinking about the privacy aspects of it. For example, is face recognition in every device the future we want? Knowing very well the information processed will be sent back to the corporation in control of the product if the opportunity is given.

  • @jimmyblade3014
    @jimmyblade3014 2 года назад

    This is a simple sorter system made from resistor banks. The resistors don't store anything. No calculations needed in the resistor banks so it's faster with less power consumption. What am I missing. Been out of the tech field for eight years.

  • @max477
    @max477 2 года назад

    If I fed AI sandra bullock and Millie Bobby Brown images the probable output is Anastasi

  • @stuff_del_emer
    @stuff_del_emer 2 года назад

    what a time to be alive.

  • @MikaelMurstam
    @MikaelMurstam 2 года назад

    Very nice video and news. I wonder what the resolution of the ADCs and DACs are and if they are low in order to hide errors and how that affect the effectiveness of the inference. The MRAM implementation in neuromorphic chips sounds like a dream. I wonder what happened to the memristor for neuromorphic chips.

  • @JonathanDecelles1990
    @JonathanDecelles1990 2 года назад

    Is the ASMR sound in the video background intentional ?

  • @LordoftheCats
    @LordoftheCats 2 года назад

    Interesting content. Love your voice! I could listen to you talk about mud. lol

  • @mikl2345
    @mikl2345 2 года назад

    Does apple neural engine work like this? Would be interesting to know how NE differs from GPU?

  • @luca21176
    @luca21176 2 года назад +1

    Amazing video super intersting as usual,It just doesn't seem real. Great news. Who taught you " Bella ciao" ?