It’s been a weekend. My wife pulled a double-shift last night, and I took advantage of the boring silence in the house to wrap up the preparations for my last week of classes, and was up way too late. For my history of evolutionary thought class, I have the laziest plan ever: the students are doing presentations all week, and are also responsible for evaluating their peers. For my intro class, I’ve got one last lecture all queued up and ready to go, and then on Thursday they get an exam. But right now, I’m really tired and should get some sleep tonight.
Now for some light entertainment. A company called Extropic is getting millions of dollars of funding on the basis of this kind of gobbledygook.
Extropic describes itself as building a
computing paradigm which harnesses the power of out-of-equilibrium thermodynamics to fundamentally merge generative AI with the physics of the world.[Extropic, 2023]
There you go. Translate, please.
I’m no physicist, but even I can see that that’s a lot of noise — they’ve garbled up a few thermodynamics buzzwords and mixed them up with the magic word “AI”. Helpfully, they go further to try and explain.
And the hardware wants to be stochastic, if we want to keep scaling. So why don’t we just cut the Gordian knot and simplify things, and implement AI algorithms in, as the stochastic physics of the world. So what we’re building at Extropic is a full stack, Physics-based computing paradigm focused on AI. And we harness the stochastic physics of electrons directly in order to instantiate probabilistic machine learning, which is the parent concept to generative AI.
Maybe I’m just tired. Maybe I’m just a lowly biologist. Maybe I’d just like to know how someone can con millions of dollars out of silicon valley venture capitalists with that kind of insipid, pretentious babble. Are VCs just not very bright?
All it needs is some goofy AI-generated ‘art’ to make it look like sci-fi…and look, the gullible tech press provides!
He who instantiates probabilistic machine learning rules the entire compost!
I don’t think I’ll ever be silly enough to spout this sort of rubbish for money. This is from their own website, extropic.ai:
Extropic assembles itself from the future
[Incoming transmission from the future]
The era of omnipresent generative AI is imminent.
Timelines have been accelerated.
The future must come to pass.
Extropic is building the ultimate substrate for generative AI in the physical world.
Building an AI supercomputer by harnessing the first principles of thermodynamics and information, like an alien would.
With this fundraising announcement, Extropic crosses a significant checkpoint in the timeline.
A milestone in wielding the techno-capital machine to birth a core technology for our civilizational trajectory.
The Extropic AI supercomputer thus begins assembling itself from the future.
[End of transmission]
After that there’s introductions of their executive team, then this bit about where their money is coming from:
Our backers
We are excited to announce our Series Seed round totaling $14.1M, led by Steve Jang and Kindred Ventures (Uber, Coinbase, Humane), with participation from Buckley Ventures (Rippling, Figma, Vercel, Mercury), HOF Capital, Julian Capital, Marque VC, OSS Capital, Valor Equity Partners, Weekend Fund, and several others.
Amongst its angel investors, we are excited to have Aidan Gomez (Cohere), Amjad Masad (Replit) Arash Ferdowsi (Dropbox), Aravind Srinivas (Perplexity), Balaji Srinivasan (Coinbase), Bryan Johnson, Chris Prucha (Notion), Farbood Nivi, Garry Tan (YC), Ivan Zhang (Cohere), Naval Ravikant, Oliver Cameron (Voyage), Packy McCormick, Scott Belsky (Adobe), Tobias Lutke (Shopify), and many more on board.
We are grateful to have such a broad base of investors committed to advancing the frontiers of computing. We look forward to accelerating the advent of our computational paradigm with this Seed capital.
One more thing…
The future is nearer than it may appear… The Extropic hardware self-assembly process has already begun:
…followed by nothing more than their crappy logo, where the name isn’t even properly aligned.
I’m guessing these are among the techbros who support Donold Trump. Either way, their actions say a lot about AMerica’s techbro-capitalist class…
Also, these technowankers seem to be associated with the so-called “effective accelerationism” movement, described here: https://www.yahoo.com/tech/effective-accelerationism-movement-doesnt-care-013934150.html
“Rather than fear, we have faith in the adaptation process and wish to accelerate this to the asymptotic limit: the technocapital singularity,” the site reads…
Well, we sure as hell wouldn’t want a technoSOCIALIST singularity, would we? Heaven forefend!
[due diligence]
https://www.extropic.ai/accelerate
[Dec 4th, 2023 / Announcement]
Old news.
I declare Extropic ISO 9000 buzzword compliant.
And yes, this news is about a year old now. So far I’m not seeing any more recent news about this. Which means both they haven’t got any results, and no one’s yet publicly asking where their “seed fund” is going.
Recent news? Depends.
Recent opinion, well, this post, and other things when I peek..
e.g. https://pivot-to-ai.com/2024/12/01/does-ai-startup-extropic-actually-do-anything/
So their business model relies on leveraging the stochastic properties of electrons to accomplish practically the same thing as a pseudorandom number generator, only at some orders of magbitude greater cost?
“When I use a word,” Humpty Dumpty said in rather a scornful tone, “it means just what I choose it to mean—neither more nor less.” “The question is,” said Alice, “whether you can make words mean so many different things.”
@8 drsteve is very close.
I am also just a lowly biologist, but I think
means they are building a random number generator based on stray voltages.
Thanks, John, I didn’t see the date on that one. And that article confirms my suspicion: they haven’t done much of anything, and their backers haven’t asked any questions yet.
[Thanks, RB]
So much for the Fermi Paradox. The explanation is either that every civilization ends up building this, or every civilization ends up going senile enough to believe this. Probably the latter.
Sounds like someone translated Depok Chopra into corporate lingo.
Not a physicist but then I doubt they are either. If they mean the first law of thermodynamics what they’re basicqlly saying is that they’ll use temperature differences and electrons to build logic gates. Something, something, end result is one of these modern things we’re pretending are artificial intelligence when they’re really just complicated blenders.
Translation: They’ll turn on a computer and run something like Midjourney or ChatGPT on it and it’ll generate heat and use electricity.
@15– I doubt their thinking is that sophisticated. I suspect it’s as basic as “AI algorithms are probabilistic; electrons are probabilistic; conflating the two is a great way to suck money out of venture capitalists; throw in some thermodynamic and singularity buzzwords to appeal to the effective altruism bros.”
microraptor: I would not be at all surprised if that’s exactly what they did. If they found lots of people believing one author’s woo/word-salad/bullshit, it would make sense for them to repackage and reuse what’s known to work.
One guy’s wrist is crushed under his atomic coffee mug and the other guy is pointing where nobody is looking, not even himself. A clear example of the AI-generated semantic failure that electron stochasticity will fix.
@16: I dunno, I think they can manage the “turn on a computer and run a program” part of it. That’s not too sophisticated even for a bunch of twits who’re all trying to make up reasons why they’re actually the smartest person in the room. Even when the whole room is filled with unqualified morons talking about things none of them understand.
I agree these guys are after low-hanging fruit though. You can get enough physics to see through this nonsense by watching one episode of NOVA. Or having even the vaugest of ideas about Heisenburg’s uncertainty principle (they seem to be suggesting they’re seeding randomness by tracking individual electrons – they definitely aren’t). I’m actually surprised they didn’t outright mention quantum computing because they sure as hell tried to imply that’s what they were doing.
I think that was supposed to have been millstone and certainly not a milestone
coauthor here!
@3: yes, it’s literally the same guy. Guillaume Verdon of Extropic is BasedBeffJezos of e/acc.
The funding was a year ago, but Verdon is notably out there still doing endless podcasts and visibly not producing and AI bubble observers were loudly wondering wtf these bozos were up to, so it’s current enough. We think they’ll run out of cash soon though.
Verdon is actually almost a physicist – bombed out of a Ph.D but that was enough to get him an actual job in quantum computing at Google – so I think he’s a real scientist who is actively trying to go full crank early.
Sure. All you need to do is to connect the logic circuits of a Bambleweeny 57 sub-meson brain to an atomic vector plotter suspended in a strong Brownian motion producer, such as a nice hot cup of tea. Easy.
Um, hello? Have you seen the crap they’ve been funding up until now?
Although to be honest, I’m not sure that the problem is that they’re not very bright per se – the problem is just what always happens when somebody has more money than sense. In this case, they have such unimaginably vast amounts of money to play with that no amount of sense could possibly suffice to keep it in check. And to make matters worse, it’s pretty much all somebody else’s money… Their incentives are all about keeping the gravy train running, no matter how silly it gets. So in a way, they’re doing the only sensible thing, within the confines of the capitalist mindset.
Well, on the one hand, quantum computing is actually a thing, so there are maybe some limits to how much you can bullshit there… And on the other, that’s somebody else’s grift: https://pivot-to-ai.com/2024/11/03/pivot-to-ai-get-in-loser-were-going-quantum/
Is there even ay evidence they wrote that text themselves, surely you can train an AI to generate the scam itself? Scamception commence!
@erik333 #23, yes that was my thought exactly. They are creating the ultimate woo generator using the large database of all past successful grifts.
Merely corroborative detail, intended to give artistic verisimilitude to an otherwise bald and unconvincing narrative. W.S Gilbert
Translation: They are going to load an AI program into a laptop, type on the keyboard and let those electrons whizz through the CPU to produce meaningless bullshit.
Dunc@22, here’s Pivot to AI’s take on Extropic. They’re not impressed.
https://pivot-to-ai.com/2024/12/01/does-ai-startup-extropic-actually-do-anything/
One guy’s wrist is crushed under his atomic coffee mug and the other guy is pointing where nobody is looking, not even himself.
Wow, thanks for catching that! It’s amazing what you miss when you’re ignoring a superfluous AI-drawn illustration! And speaking of which, WTF are “Thermodynamic tronpriteirs”? That might make a good name for a jazz-fusion album (“fusion”, get it?)…
“Out of equilibrium thermodynamics” is perhaps the emptiest buzz-phrase I’ve ever seen when dealing with computers. Computers get hot. Server rooms get very hot. That’s out of equilibrium already. Essentially, they’re bragging that they use so much electricity and produce so much waste heat that they’re doing the equivalent of burning down the amazon to power their instance of ChatGPT, which they will use to run physics calculations that could be done much more efficiently and quickly (and … correctly? Who cares about that, it’s AI!) using dedicated and validated physics software.
Flash over substance, and buzz-words over actual correct results.
Remember, according to ChatGPT there are four Es in the word elephant. We’re going to use this for real work?
Pretty sure they got AI to write this marketing drivel.
In a previous century, C&E News (I think) published a buzzword generator. Three columns. Adjective, adjective, noun. Pick a word from each column and Bob’s your uncle!
strategic scalable network (for example)
Well, of course we programmed the list, activated by a robust, stochastic generator and it spewed out the most delightful, satisfying malarkey one could hope for. As a prank, we stitched together the “best” combinations and submitted it to our supervisor as a prank.
Alas, Colin, our boss, an Oxford graduate, saw through the ruse immediately. However, always getting the last word, he penned a lovely gobbledygook reply using his vast vocabulary alone.
Google “corporate bs generator.”
Raging Bee. It’s not “Thermodynamic tronpriteirs”. It’s “Thermodymmic”. Whatever that is.
I think PZ may be on to something here…
…the most obvious being that the VCs themselves operate on two rather interesting twists on thermodynamics:
• VCs assume, in substance, that their role is to open closed “econodynamic” systems by providing an economic input equivalent to providing more heat. Of course, this is all largely illusory, because all it does is change the definition of “the size of the system” — the only way that Maxwell’s Demon can be valid is to treat a subsystem as if it’s the entire system, which then successfully/rhetorically ignores frictional/inefficiency effects on the larger closed system by pretending it doesn’t exist as a constraint. (This is one of the implications of Szilard’s lectures on applying the Second Law to information theory… three-quarters of a century ago.)
• BTW, Maxwell’s Demon deserves to be personally compensated handsomely (that’s the “profit motive” at work), and VCs themselves are Maxwell’s Demon. Go ahead, look at some of the academic literature that rationalizes arbitrageurs as having outsized contributions to “economic efficiency,” and consider that what VCs are exploiting is an inefficient imbalance of available information rather than of available capital.
For me, the most amusing thing is that their super-advanced semi-transparent holographic display is facing away from them.
But given that they both appear to be staring into the abyssal void, rather than looking at anything in their own universe, it probably doesn’t matter that much
As a computer scientist, this is all gobbledygook.
It almost sounds like they are saying that they’re going to have random errors in their chips like it’s a good thing? There are hardware random sources, but they have to be carefully analyzed to be drawing from a particular distribution to be used; any bias and you’ve lost the guarantees of randomness.
“Out-of-equilibrium thermodynamics”—I assume they meant non-equilibrium TD? That’s the standard term for it, anyway. I can kind of understand how they’re trying to use it here. “non-equilibrium” describes virtually every real world TD system since equilibrium since things in the physical world are constantly undergoing changes in state/energy transfers, and the distribution of energy tends to not be uniform in said systems. The calculations in non-equilibrium TD are a lot more complicated since they’re basically dynamical systems where you have to consider a ton of variables/parameters changing over time. They’re a lot more susceptible to fluctuations, and it’s difficult to describe them in meaningful macroscopic terms—everything is filtered from the bottom up by the individual “units” in the system.
Certain subfields of TD (like statistical mechanics) deal with these units by describing them probabilistically (as is usually the case in quantum mechanics)…I guess that’s what they were going for with the whole “stochastic physics of electrons” thing? It’s still pretty vague.
“And the hardware wants to be stochastic”? Whatever they’re trying to get at here, it’s a really fucking stupid way of phrasing it. Is it meant to refer to physical behavior modeled by probability distributions? Talking about stochastic processes in terms of AI makes more sense since probability-based models are a staple there (the Bayesian framework, stochastic neural nets and whatnot).
Anyway, the idea of using this stuff specifically for generative AI seems really pointless to me since we already have a bunch of paradigms that are more straightforward and less expensive to set up. They’d get a lot more leverage out of it if they were using it in a more scientific sense to computationally model IRL systems. Since we’re talking about AI, neuroscience could get a lot out of it.
Take heart PZ. I think that biology is the premier science when it comes to long impressive terms, it’s all Greek to me! I am sure you can con millions out of silicon valley if you wanted to! And yes, venture capitalists be stupid!
Whenever I see something like this I think they should be accompanied by a ten year old published paper describing in principle the thing they are trying to do. Way too many people think that progress comes from throwing money at random assholes with a good sales pitch, rather than being the final product of a deep, blue sky research program.
“Thermodynamic tronpiteirs”?!
@ #34 Karellen
They would see the display, but it would read backwards. Their goggles could have an AI chip that reversed the view.
Also, that instrument the AI-generated Black guy in a lab coat is holding looks like a needle one might use to sew up with string a beef roast or turkey. The AI-generated caption should explain it.
This is amusingly similar, to my mind: