This came up on my YouTube rec’s and there’s a text article to go with it, and of course from there, I got other stuff on the topic, because the algorithm is a merciless thing.

To set this straight, there are reasons this looks like the beginnings of the next bubble:

1. The idea a company nobody has heard of (and doesn’t have a Wikipedia page) has solved a seemingly fundamental problem – the fact that analog chips are imprecise – is at best a half-truth. Analog chips aside from their individual imperfections are sensitive to noise in their environment. They will never actually act perfectly like a digital chip. More likely, it is utter bull, or uses some sort of rounding on the output circuit, but it doesn’t matter because…
2. Analog devices work just fine the way they are, aren’t in any way new, and despite the marketing hype here, were at general-purpose levels by the 1960s. It was only in the 1980s that digital chips became competitive. This isn’t even the first time hybrid systems have been brought to market. So treating this like a miracle is disingenuous. At most, they’re making them accessible to modern languages and miniaturizing them. All good things, even impressive things, but let’s meter our expectations here; if the idea is not to in any way replace digital CPUs, and only act as first-line processing to hand off a wake-up call to a digital CPU, the tasks are going to be in a much more limited scope than they’re letting on.
3. The claim here is of course that it’s the magic bullet that will make A.I. more energy-efficient, which is using the old buzzword to sell the new buzzword. Tech bubbles are nothing if not parasitic, off each other as much as gawking investors and breathless press. NFTs piggybacked off blockchain; The Metaverse took the NFT concept of digital scarcity and applied it to fake real estate; Web3 and The Metaverse were like dogs circling sniffing each other’s butts; and Web3 ultimately wanted to make itself the MCU of all of it, which makes it all the more hilarious that it all came crashing down at the same time, only for A.I. to have quietly been building itself up in the wings for just that moment to burst onto center stage. The grifters are not stupid; they’ll sell the new snake oil using the old snake oil as soon as the old snake oil starts looking stale, and the next one is always in the wings. A good conman loves a stupid mark and always knows when it’s time to move on. The primary difference here is making the jump from software (easy to promise on and never deliver) to hardware (which needs to look like it works in a room).
4. This got pushed to me unprompted after looking up probabilistic computing, which tells me that it’s being pushed to anyone looking at computer stuff. Probabilistic computing has essentially nothing to do with analog computing except for the use of analog flash memory as part of the hardware. That and the next recommended video is 2 years old, with a clickbait title, which of course means that like A.I., this one is emerging from its cocoon after having some time in the oven right as the last one is showing signs of falling apart. The plural of "anecdote" is not "evidence," but it’s honestly either this or probabilistic computing and probabilistic computing is still squarely rooted in academia at the moment and the technologies are still sitting at proof of concept rather than being at final scale and in the hands of grifters with dollar signs in their eyes. It wouldn’t surprise me if probabilistic computing isn’t the next one after this, though. All it really needs is an unscrupulous face who can claim it’s magic.
5. Speaking of, that dude has the expression of a con artist who just saw his mark take the bait hook, line, and sinker. There is nothing to say his little speaker device isn’t sending some other signal to light up the LEDs on the chip device, or that the device isn’t just listening for a specific frequency, because the noise is 100% fake, and well outside of the range that most humans are able to easily produce, though I will say that as a guy with a brother, yes, there are ways of doing it, and we found them, because that’s just what brothers do in their unending quests to annoy and one-up each other.

There are, however, reasons to be hopeful, because:

1. IBM is doing similar things and is a company everyone has heard of.
2. As mentioned, whether they fixed the imprecision or not is irrelevant.
3. Also as mentioned, none of these ideas or technologies are new, meaning this one is perfectly achievable, and if all that’s really happening here is making them accessible to modern programming languages and throwing them at sensors in novel ways, this one actually has a use case, which is more than any other tech bubble in the past decade can say. The only question is whether it will get there before the hype train sucks all the oxygen out of the room like generative A.I. did for actual research. If I had to pull out my crystal ball, I’d say that this is going to be a lot more like The Metaverse where being a thing that already exists is going to make it a much shorter hype cycle, because tech bubbles always inflate biggest when people can’t look to their left and see they already have what’s being sold. Once the dust settles, we might actually have something useful out of it.

Being realistic, the idea that there might be real potential to slow our electricity usage after blockchain single-handedly undid a decade of green energy efforts and generative A.I. somehow used even more energy than that is, shall we say, highly attractive. This is one I unironically want to work out.

Analog-to-digital converters (ADCs) take up most of the energy of something like a telephone, and solving that in 2017 involved cutting out the digital component entirely. Which is to say, yes, there is truth to the sales pitch here. When you’re talking about sensors, keeping things analog, inasmuch as analog systems that stay analog bypass the ADC overhead, might legitimately allow for carefully limited chipsets to save energy as is being suggested, though whether a fully programmable general-purpose CPU shares that benefit, I don’t know enough to say.

Now, I don’t think this is going to actually solve the problems of A.I. or its energy usage because, to be blunt, this has nothing to do with generative A.I. Ultimately, generative A.I. is what’s sucking entire countries’ worth of electricity to train its models and another full charge of a cell phone’s worth every time someone uses it. "A.I." here is a buzzword. Being realistic, analytical A.I. is specifically what’s in question here, which is the significantly less sexy, but infinitely more useful kind that nobody is talking about because it’s not useful to mediocre white men who have no creative skills or morals that might stop them from stealing other people’s work. Analytical models are great for stuff like weather prediction or motion sensing and when it comes to sensor input, that’s perfectly valid. Unlike generative models, they don’t need a whole lot of power or fancy hardware and one might call them more of an algorithm depending on how deep they need to be. Real talk, I have analytical models in a number of devices in my home. They’re weather clocks and after a brief training period, they are surprisingly accurate at predicting what the weather will be like for the next 12-24 hours. This staggering technology can be yours for all of $35. Which is to say that for basic crap like the sensors on your phone, it’s perfectly reasonable to take a solved problem and try to offload it onto hardware that can handle the algorithm without the expensive ADC operation.

This is ultimately not going to be the computing revolution it’s being sold as and it’s kind of hilarious to listen to them talk about modest use cases like saving your battery like it’s earth-shaking, but the point of a hype bubble is to get people excited so they give you their money. They’re being shockingly realistic here and the honest truth is that exactly nobody is going to notice in practice, provided the solution hits market.

But as for a solution that can hit market, I’d honestly like to see it. Moore’s Law is dead as we know it and if that can let us focus on bringing other technologies into the modern era for their benefits, I’m for it. We can absolutely expect to see it being advertised for things it has absolutely no business being used for, but the trick here is that they’ll need the right people to commit to it. Hardware companies are a completely different market than software companies and there are far fewer of them, which means that demand isn’t going to come from a board room; it’s going to be pushed directly by consumers if it gets off the ground. You’ll see analog security cams, analog burglar alarms, analog thermostats, and analog IoT devices before you’ll ever see it in an iPhone. This stuff is going to get marketed to an upper-middle-class demographic and whether it works or not, if people think it does, then you’ll see demand for it to be in everything. The difference here is provided it works, it’s not going to be a bad thing. "Either you die a hero or live long enough to become the villain" absolutely applies here; if it actually works and ends up in everything fulfilling reasonable use cases, it’s going to quickly become boring. It’s only if it doesn’t actually deliver that the bubble will burst rather than slowly deflate.

I’d like to see a world that uses less power. I’d like to see a world where computers that can’t get any smaller start getting creative with tech that takes them just a bit sideways, stuff like analog to save power and stuff like probabilistic circuits that allow them to be faster and more efficient at certain tasks. Creativity can come from how to use the noise inherent to the technology for things like true randomness or smarter pathfinding or just better calculators. I’m old enough to have heard a teacher say "you won’t have a calculator on you everywhere you go."

The only question here is whether the practical benefits, if they really work, will be overshadowed by whatever magic someone might try to sell the credulous once this gets off the ground.