The Next Era of Computing | Extropic
ฝัง
- เผยแพร่เมื่อ 26 เม.ย. 2024
- Exclusive first look inside Extropic, pioneers of thermodynamic AI compute, on episode 41 of S³
Startup: www.extropic.ai/
Blog ✍️
saturdaystartups.beehiiv.com/
Socials 📱
/ jasonjoyride 🐦 | / jasonjoyride 📸 | / jason-carman-63b384199 👨🏼💼 | www.jasoncarman.com/s3 💻
FIRST 4K S³, FIRST THERMODYNAMIC CHIP! (Full interview with Guillaume is here: th-cam.com/video/QjVOfM2EBnE/w-d-xo.html)
@eblman5218don't trust that nerd he denied the threats of AGI not being superaligned
@eblman5218 do you not know what the alighnment problem is?
That's would be so funny to see Quebec revolutionizing the computing technology
They won’t
I’m a computer engineer, trained in the US and work in lithography. Highly doubt they will produce anything more powerful (with less error) than classical computational methods in the relative near future. If this can match performance of classical computing, maybe it’s viable, but they haven’t really proved anything as a company to validate the founder’s claims. I’ve looked at what they’ve published.
Frankly, best bet right now for non-photonic QC is undoubtedly IBM.
@@Laminar-Flow it was a joke but yea
The next big thing will be LPUs and photonic chips ¯\_(ツ)_/¯
@@dranon0o Photonic quantum computers are highly unlikely to beat out superconducting quantum computers to be honest.. IBM and Google both use superconducting qubits and IBM is frankly the world leader right now. That’s to this companies credit because they are going to superconducting route, but there’s still problems with their claims. Also, to be frank, they aren’t going to solve superconducting qubit error correction before IBM or Google. Odds are just simply against them when both companies have thousands of physicists and engineers and access to some of the planets most advanced semiconductor fab technology. The issue with photonic chips are the error correction is significantly harder than with superconducting ones... It’s, in essence, an impossibly hard physics problem to solve and will require far more than a couple dudes in a lab to figure out as far as science knows given the amount of investment going into that problem..
LPU’s aren’t going to take advantage of economies of scale remotely similar to GPU’s in the near future. GPU’s in terms of Flop/s are pretty much unbeatable for their price and are a generalized parallel processor that can be mass produced with massively high complexity and used to solve most problems, not just one. Maybe there’s a place for LPU in future hardware but the GPU isn’t going anywhere and it’s unlikely making a dedicated LPU in addition to GPU will be a common practice. My desktop GPU right now has 76.3 billion transistors- it’s highly unlikely to be outdone in terms of both economies of scale and one other thing- In terms of LPU/TPU/GPU the shared issue limiting them all is in the time it takes you load a floating point into the processor from memory; the processor could have completed hundreds thousands of floating point operations. There are fundamental limitations that affect all computing hardware, superconducting processors included. Which is made apparent by the fact that this founder doesn’t have education in computer engineering, only physics. Groq LPU is impressive but still isn’t remotely as effective as a large array of classical parallel processors such as Nvidia’s Grace architecture..
I mean yea tbh it would be funny to see a Canadian company revolutionize the industry the US gov and US companies are spending literal billions to figure out. But it would also be highly unlikely (I figured it was a joke).
@@Laminar-Flow Oh, you're that guy
@@fosatech Yes, I’m honest my friend. Cool radios.
Beff Jezos the man
Just checked it's bio and it's really him!
Funny he uses 'mids' in his latest tweet.
Most dont know mids is a term derived from my industry, 😅 yes, the herb.
Reckless fool, that's who he is.
7:02 "and of course,
open sourcing our software" music to my ears.
These guys are the real deal!
Can't wait for the interview Jason❤
So instead of using 0s and 1s, they are using noise generated from heat as the form of computation? That goes way over my head but it sounds cool!
Makes me think, are we looking at DNA all wrong ... is there a totally different way of reading it and our surroundings ???
It’s not replacing computation right ? It’s just replacing the way of generating random values ?
@@hugodemenez I think you’re right
@@hugodemenez @Settiis I think you're missing 1 thing... it's fraud until it works.
man as an aspiring engineer, this channel is awesome, all the new people. companies, inventions that will shape the world. just keep going! this is a million kind sub channel!
the science and their tech is so far out, it barely makes any sense to laypersons like myself. a longer episode w/ an in-depth interview with Gill would be really cool
Great thing we do that every day after the main episode! (Interview is now live: th-cam.com/video/QjVOfM2EBnE/w-d-xo.html)
100% I want MORE!!!!
@@s3_buildfancy eye rolling someone who wanted more of your content, as if they should know your content releasing process. Jerk.
@@s3_buildFor some reason I had no idea those longer videos existed either, even though I have notifications on. Literally found out just now.
Maybe dont overthink about an emoji 😉 @@joshismyhandle
Thank you, Jason, for S3!
Very excited to see how far this goes!!
Most underrated channel on TH-cam. Thanks for putting this out there
I love this guy. Saw him first on the Lex Fridman podcast. He's building a really nice fridge here.
"Scale civilization to the Stars" we/humanity need visionary thinkers like this in places of government and business leadership to positively impact society. Please keep pushing and no, you are definitely not going crazy!
Not human civilization though. Did you know his positions on the human extinction from AI? On short, he welcomes our demise.
Very cool, and kudos to the team
Awesome footage and production 🤩
This is fantastically exciting! Thermodynamic computing always made the most sense to me, so I'm happy to see an ambitious startup on this path. Good luck!
Loving S3 content criminally underrated
This episode is a huge milestone for your work and for e/acc. Big congrats!
This felt like the docking scene from interstellar
Docking 🤤
Because its fake therefore it seems like a movie
holy hell what a compliment, thank you.
@@dumb8671 I've heard conflicting things about the legitimacy of this project, could you briefly explain why it is fake?
@@zrakonthekrakon494Yeah I just open your eyes you know actors in commercials they poses doctors lawyers but we all know their actors you know it’s like the same thing the music’s played throughout it’s just that the contraptions aren’t really like a computer just everything doesn’t add up you just kind of open your eyes man
Great video! Hope your channel grows and many more people start loving technology and humanity again
i already own so much noise, i am ready to start! 🙂
Very mind blowing demonstration. many thanks!
great job boys
Superb. Amazing.
So excited about Extropic. I wish I was able to work in the US and join them. Great video 🤩
You had me at "open source"
This channel is going to blow up
wow this is amazing, i feel like you guys have cracked the code
À Sherbrooke en plus, très cool comme projet, lâchez pas!
Had the same idea a while back.
This is the way ...
for building AGI.
Using nature to compute like an analog computer.
Didn't realise it would be practical but someone actual made it
bravo sir, bravo
LFG, love your guys videos.
great pro video!
Digging your videos man.
amazing
Excited for the deep dive... love all the tech you've featured on the channel I just can't for the life of my understand what he's talking about 😆 Hoping you can bring in some real world examples with how things are done / were done, then how this would do that same thing in a better / more efficient way, in a way that doesn't sound so nebulous
Some of these videos need an eli5 section. This one especially.
Yeah, exactly why is this a big deal? How will this transform human lives?
I think that's because they don't really have a concrete idea of what their system can do, even if they created one, because they haven't done it yet.
I don't know what I'm talking about but: computer chip sets use transistors to take electricity signals and make them less wobbly (noise and heat) and magnify them or turn them into a different kind of signal. We're hitting the limit on how small we can make transistors because the electrons in the signal to them are basically wobbling and leaking out of the transistor when they are changing and it's not really possible to stop them. This technology is proposing that it is possible to use the electrons properties that cause them to wobble and leak to drive chip functions and get around the that barrier.
simple explanation:
increasing demands of AI means we need better chips. but typically to make chips more powerful, you need to make their components smaller. when the components are very small, things like vibrations and random disturbances become noticeable because you're operating on such tiny scales. Extropic want to harness this "inherent noise", to make even more powerful chips. this makes sense (on paper) because AI fundamentally requires irregularities to work properly (it is a "probabalistic" system, rather than "deterministic"). Up until now chips have tried to avoid these effects, so Extropic's line of reasoning can be summed up as "why not use the inherent irregularities of nature instead of trying to avoid them", thereby opening the door to way more powerful chips.
There's lots more detail ofc, but this is the general thesis
The newsletter is good for this
This video makes me want to cry tears of joy
Great passion from a scientist entrepreneur‼️
beff is doing god's work ⚡
I loooooove how you look at things from a massive fundamentals approach - using electrons to mimic generative AI physically!??!?!?!?! Materials science + AI is the most valuable role in the world right now
This legend living his destiny! Just wow!
It won't replace classical computing. You don't need a QC to change traffic lights. What it will do is solve problems that are impossible to solve with classical computers.
This is either going to be massive or they are talking nonsense just to
attract investors
Subscribed.
🫡 Godspeed from Baku!
Does anyone know what song is playing at the very beginning?
Woah! At first i was spektical now iam even more
What is the price tag on the fancy fridge with next level IoT?
guilaume is just radiating his passion for advancing the frontier of compute!
Moving beyond from known to embracing unknown.
How do you scale quantum generative ai, open source or not? A cryogenic datacenter?
So, any numbers? I'd like to know how more efficient it is compared to current computation methods.
Is it as every efficient as the brain?
This is a PR piece I wouldn’t expect numbers or scaling this tech any time soon
Thank you Jason Carman for sharing startup's stories on Saturdays (S³) from Sherbrooke, Quebec, Canada and presenting how the CEO and Co-Founder of EXTROPIC (Guillaume Verdon and Trevor McCourt) Entrepreneurs and Scientists are contributing to the future of AI and Quantum Computing. I learnt a new expression... "Thermodynamic Computing".
Where can I buy your stock?
Mad scientist! Hope it all works out! Everything peaks and falls at some point just hope our point isnt tooooo soooon
Nice to see the next shitty theranos startup is here.
Is your channel by any chance named after the S²?
Please add startups videos to the playlist and even podcast thank you 🙏
Wait are you trying to run regression analysis on raw nature?
feels like Alan turing's lab of the 2024 👍🏻
I need an ELI5 section for this kind of topic
This is what a Start-up looks like.
So, the future of computing is based on quantum computing, Josephson junctions, and a dilution refrigerator? Color me skeptical.
Ok
how old is this field
Quantum computers have yet to be shown to be faster than standard computers at ANY task.
Every single paper that has come out that has claimed to be faster at performing tasks has been shown to be able to perform as fast or faster on a standard architecture.
Interesting
Does anyone know how to invest in this startup?
But can it play Crisis? As a gamer and future scientist I really like where Jason took this narrative.
❤❤❤❤❤
1. Choose a Universal QC Gate-set.
2. Use their tech to creat the specific distribution of each Universal Gate.
3. Expand your algorithm using this Universal Gate-set.
4. Run your new application on their Thermo’puter.
Could you build this into neutrosofic logic???
How can I invest? 👀
I recently had this idea that in order to make ai efficable on a large mainstream scale, really just agi, we would need to reinvent the transistor to be the neuron or node in the neural net whereas right now we have many transitiors pseudo-emulating each node. Sounds like this is what these guys are doing but in a way much more sophisticated than I imagined
All change is computation. That which increases entropy is computation. The universe computes reality nearly for free.
So the question becomes: how low can we go? Can we do away with our wasteful/limited computational abstractions and simply ride the wave?
Be careful, remember Wardenclyffe Tower
Quote: "Explain to me as if I was a two-year old"; what is, or was, Wardenclyffe Tower? 🙂
@@benaiah1960 A giant structure built by Nicola Tesla with the goal of providing wireless power everywhere across the globe. He sunk all of his money into it, it didn’t work, and it bankrupted him.
It’s also the title of a fusion album by Allan Holdsworth.
Awe-some
Just .... Holy FU$K!
Seems interesting, but also very impractical to scale and deploy at a large scale. May be a room temperature superconductor (when discovered) will make this much more feasible. But then again, a room temperature superconductor will revolutionize the whole world, not just this :)
Man I really really hope this isn't gonna be one of those revolutionary fields of research with no practical applications
Normand Marineau
Ai jobloss or extinction? Maybe. But with swell robotics everywhere, Ai jobloss is the only thing I worry about anymore. Anyone else feel the same? Should we cease Ai?
We are reaching the cliff of ASI
is this guy @BasedBeffJezos
Guill is an amazing person! I love Elon and Guill! May thermodynamic God be with them all the way through!
Do people highly educated in thermodynamic physics know of a higher consideration of nature that overrides accepting that diodes can rectify Johnson Nyquest thermal noise and, given the orderlyness of consistent orientation in parallel, aggregate a DC residue from each diode into power at any scale?
Hello Mr simple storage service
I don't wanna go to the stars.
why does this feels like a parody? 😂
His beard is grandiose
I did not understand one single thing that was said. Is it just me?
Nope. I don't have any freaking idea either 😆
Same here... But this one sounds very cool...
The reason is because the video's title should have included the phrase "Thermodynamic Computing"
Cuz they didn’t explain it at all
@@SGN30 Thank god🤣 Apparently S3 is gonna realease an interview with the guy. Let's hope it's less buzzwords in that video.
holy s$@!
based
We have enough intelligence; lack of intelligence is not our problem. Our problem is for what purpose we use this intelligence. We can have intelligence to build a bridge, but why do we build that bridge needs a reason that cannot be dictated by mere intelligence.
These people who are going to the stars, need therapy, but no amount of Ai will get them to therapy.
You sound like the priest who burned so-called witches in the past because they worked on chemistry and medicine.
@@ayo2242-go6hw He's making an observation, not a condemnation, I agree improving technology, especially computing is the most important frontier of advancement right now, but it still can't answer why we should build bridges
Wobble wobble
Smells like series B
a team of remote viewers I follow looked a few years into the future.. I think 2027.. and saw chips that functioned more biologically and they said they felt organic and more like brains
it's beff
Sounds like an Asimov positronic brain.
Seems like a lot of vudu magic and and snake oil to me. I think there’s just not enough to say on the topic right now for a video to be helpful.
Same. I think this is a parody. 😂
Yep the whole video is shot like a proper startup pitch but I just can't shake the feeling that it's a mock video, like when students learn to shoot and edit a startup pitch and their pitch might as well be reading lorem ipsum.
The only thing they've shown that they themselves have to show for was the jpeg on the computer screens over and over. They don't own the quantum computer, it's just something flashy and well known to lend them the impression of legitimacy. Remove anything that they don't own from the video then they pretty much have shown nothing at all. All talk and no substance.
I wouldn't give them any money.
Jealousy
With general purpose extropic DC - electrical - power - coproduced - with - refrigeration - devices the requirement for a mature civilization is very intense.
The science fiction novel Tau Zero teaches that sometimes the way out is more. In the novel the starship had ro step closer to the speed of lght to survive.
Open sourse development as desired by you, me, some AI & computer developers, and many others may balance social issues with exuberance everywhere. For example economic refugees in LA could use security devices including radios and witness cameras to be protected from street extortionists.
Extropy is an underused general concept. There should be many ways to fabricate DC - power - coproduced - with refrigeration - devices
Another method to plausibly transform ambient heat into electricity with equivalent cooling essentally consists of two electrodes closely face to face (~1 micrometer) in a vacuum wired to an external electrical load. The face of the [Emitter] electrode is covered with a uniform array of LaB6 tipped small diameter carbon nanotubes grown straight out. The face of the [Absorber] electrode is covered with small scale graphine flake char. [Rice U 2014]
Thermal energy mobilized unattached electrons will tend to free themselves outward from the emitter tips and drift at ~1 million meters / second @ 25 millivolts (thermal electron energy @ 20 C) to the absorber which tends to collect them.
A negative charge accumulates on the absorber. This repels oncoming electrons slowing their forward drift, cooling them. The absorber electrode charge is simultaneously the repelling cooling and the external electrical load voltage. The drift current and external wire route current are the same. The DC electrical power consumed by the electrical load depends on the load resistance. Thermal energy absorption always equals the electrical yield.
Wire resistance is a practical loss not a true loss so lt is overcome by added device output. Extra cooling thenalances the heat given off by the wire loss. The performance of the device is expected to be modest in the beginning but improve rapidly. Even early devices are expected to last a long time. There is little place for obsolence if the first installed device works adequately. They will withstand being short circuited indefinately up to an electromigration limit.
Here is an impractical device that is easy check for mechanical workability. Its parts are large enough to act as everyday mechanisms but small enough to work well with the nanometer scale thermal motions of gas molecules. This device hypothetically creates self powered thermal diversification:
Sketch made with keyboard characters:
COLD tank ())--:WALL:-->> HOT tank
Key
()) = Paddlewheel.
-- = Axle. (Continuous from end to end)
: : = Axle tunnel going through a wall.
>> = Lumped friction element
Please visualize two tanks full of air separated by a very thin wall that allows the rooms to hold their heat independently with minor leakage through the wall. The wall is thin to delicately support billions of separate nanometer scale short axles running straight through loosely enough to rotate freely but not leak very much heat so the rooms can hold separate temperatures.
On the left side, a very small paddlewheel is mounted at the left end of each axle. On the right side, lumped friction elements are mounted stationary in place on the wall, one for each axle, for the right end of each axle to run through. The lumped friction elements convert the mechanical rotation of their axle into heat. The lumped friction elements do not impart Brownian motion to their axle.
Brownian motion (a nanometer scale effect) turns the paddlewheels at random speeds randomly clockwise or counterclockwise. This random rotation is turned into heat by the lumped friction elements.
The committed, linked, and functional roles of the walls, paddlewheels, axles, and lumped friction elements in differnt places should systemically produce a divergence in the thermal energy in the two tanks without adding external energy.
The analogous conversion of thermal random motion of clean gas atoms or molecules into macroscopic fluidic power by aggregated aligned very small tesla vavular conduit fluidic rectifiers may also be possible.
In ~1980 the ~solid state device lab at the University of Virginia Charlottesville made THz rectifiers, one use being detectors for radio astronomy. They knew they needed to make them small for femto farad
Junction capacitance. They couldn't register well at the time so they made open face diode array patches so they could find them and select one diode to wire up. The diodes were Au electrodeposits in SiO2 pores abutting N type GaAs. I bought one and had it sent to a lab in California. They put some conductive paste and a lead on top and tested it. It made a clear output under professional conditions. They went out of business and never sent me the chip. The Virginia Lab spun off Virginia Diodes Inc.
There may be ballistic rectification between carbon nanotubes embedded in an orthogonal array abutting a graphine sheet.
Au/InSb diode arrays may be worth considering too.
ALOHA
Oh so it still needs to be super cooled to work? Sux that u cant run your chip at room temperature.
A few questions come to mind.
Hmm what amount of legacy infrastructure is this chip equivalent to? How many h100's does it replace?
You need a room full of white dude phd's to babysit your chip. Can this realistically scale?
This isn't meant to run the latest games on your custom PC, but rather to help run scientific simulations and algorithms that take classical computers millions of years. It's impractical in the same way current supercomputers are impractical.