It feels like we are seeing the result of diminishing returns with technology advancement.
To make a product that Apple believes is significantly better than the competition they had to design a very intricate solution that includes:
* High end look and feel not similar to the bulk of their products with lots of textile webbing and memory foam ear cups. These get way more wear and tear than normal electronics being exposed to sweat, sunscreen, etc. On top of that these must be safe for long skin exposure and comfortable across many head shapes and sizes.
* High quality magnets with custom speaker design for low THD and large frequency range.
* 2 custom ASICs built for sound processing and low power bluetooth and 10 audio cores each.
* 10 microphone array for ANC and wind noise cancelation.
* Multiple accelerometers + gyros head tracking with spatial audio.
It you remove one of those components, I'd be surprised if Apple still shipped this. At least 40 people worked on designing and engineering this new headphone. There are not many companies in the world with the right kind of talent for this and Apple happens to be one of them.
I don't understand the point you are trying to make. You lead with a statement that sounds like disappointment, but then build up to the engineering complexity involved. What are you trying to say? That this doesn't add up to a great product? How long have you been using them for?
I think the parent commenter is saying that in order to make a superior product these days, you need an extraordinary amount of investment for R&D which can only be made up by selling it at volumes that someone like Apple can achieve. I think their comment is about bang for your engineering buck.
This seems like a win for everybody. In the past a lot of people spent money on luxury goods which translated to over-paying brand name and status symbol. I would consider products like Beats headphones to be in that category; they were technically inferior to many other headphones but cost as much or more than professional headphones. It is interesting to see Apple engineering maintain broad appeal in product design while adding quite a bit of functionality.
I think Metcalfe's Law [0] for networked services is a bigger driver for winner-take-all, than needing deep pockets to compete in a mature product category. For consumer products, a small number of large players can own entire market segments between them, but still compete with one another, as we see in other capital-intensive industries like cars or smartphones. (And it's fair to see that as its own problem; but it generally devolves to the question "if markets are universally the most efficient form of exchange, why are firms a thing?" [1].)
Not necessarily, especially in areas like sound where it is essentially ‘solved’
Why should humanity put so many resources towards improvements that no mere mortal can notice? It’s probably better to put the talent on other problems...
> I think the parent commenter is saying that in order to make a superior product these days, you need an extraordinary amount of investment for R&D which can only be made up by selling it at volumes that someone like Apple can achieve.
True if you're selling average products for average people. Not true if you have a niche.
I don't know how you read a negative or judgmental tone into that comment... GP is basically saying that consumer technology has advanced to the point where you need large teams doing complicated or deeply technical (and somewhat incremental) improvements on existing technologies.
It's not a judgment on the product. It's just a lament that the days of individuals or small companies innovating in the consumer electronics space are (long?) over.
I didn't see it as a judgement, just didn't understand the point. I guess you have provided a little more clarity. How sad is it to generalize the idea that disruption can no longer occur. That is one of the saddest things I've read all day. There are numerous cases where companies 'advance technology' into a product to the point where someone comes along with something simpler to undercut the existing player. Apple isn't a good example of this because it's very hard to disrupt their products. They have extremely focus on the end user and simplicity.
It's not just the values they place on the end user and simplicity, it's their huge amount of cash they can use to test and deploy technologies (think spatial audio partners) and their tight grip on supply chains (try sourcing some of the same materials an iPhone is made of).
I mentioned elsewhere in the thread, the best chance at making a wave as a consumer product concern is to make waves on some sub-component of the larger product being produced by the majors. You aren't going to compete with Apple, Sony or Bose directly, but you can improve on new technologies or extant processes and hope to get subsumed. That's the truly sad part.
No. Apple needs large teams doing complicated or deeply technical things. That's their market. They're in consumer electronics. If it takes 50 people to ship 100M units, then it takes 50 people.
The first sentence only experiencing diminishing returns can be read negative. Also everything is easier/cheaper/faster than ever for individuals and small companies to enter the hardware space.
I suspect the best path for success that individuals or small companies can take right now is to focus on one piece of the larger product and aim for acquisition or licensing, sort of like the pharmaceutical space.
I think he means headphone tech is already complex and they needed to raise complexity a lot, lets say 10x to get something that maybe has 1.2x better sound. At least that's what I understood. In other words, you get diminishing returns.
You sound like you are selling these headphones. Yeah, making wireless headphones with active noise cancellation is complicated, what else is new. Same as making a smartphone, or a laptop, or a car. "Significantly better than the competition"? Sure they are if you listen to (or are) Tim Cook or some Apple fanboy, but really buying an iPhone is mostly a matter of preference, not really being "better than the competition". And "preference" is the best case scenario, because quite more often that's a matter of being victim of aggressive marketing.
I mean, AirPods Max may be actually better than the competition (unlike regular AirPods, which are a bestseller anyway). But they were just announced, why the hell would you assume that they are any good at all?
> But they were just announced, why the hell would you assume that they are any good at all?
I'm not an Apple fanboy, but people said the exact same thing about the M1, which basically blew everyone away. So the track record pretty heavily favors Apple here.
Because Apple released a product recently that received excellent reviews, all Apple products are assumed to be good? All Apple products come out of the unit that produces their processors. All Apple products are owned by the teams responsible for their Macbook / Airbook lines. All Apple products can achieve whatever innovations that occurred with the M1 chip.
I understood what you're trying to say. I still disagree.
The products and technologies involved are too different to take any goodwill and apply it universally. Even in the same product there are issues with subcomponents: The M1 is great, but the touchbar is still mostly useless and the keyboard only improved recently. How can we possibly say anything across products?
Even though they were not as recent, I believe your argument would be more compelling if you used Apple's more similar products, e.g. AirPods Pro, or perhaps the latest watch.
I balked at the price but if they've solved the wind noise problem I'd think about getting it. I love to walk outside when I have calls but the wind makes it impossible most days.
Google colab is a much better tool than a new computer for getting into ML. It's free, requires no dependency management, easy to share, and has tons of example notebooks you can reproduce as easily as duplicating a google doc.
Not to mention a Linux based workstation in the limit will have fewer headaches than mac these days. Package management doesn't require homebrew or dockerized everything, selinux is surprisingly easier to configure than the Mac security subsystems, etc.
I don’t know. As someone who recently built a Linux workstation but whose computing is generally done on a Mac, I had all sorts of trouble with surprising things. Like my mouse stopped working one day ... no idea what happened, tried googling, tried asking, but nothing I did fixed it. Ended up just reinstalling the OS.
Having had a similar experience, my new strategy is to set up a server and just ssh in to do ML development. That way I don't have to worry about mouse/display/wifi drivers.
You can make a model of any size with deep learning.
If your concerns are about over fitting there are lots of regularization techniques used in practice like dropout, weight decay, and data augmentation.
There's nothing preventing you from sharing weights across layers, and would be interesting to see some research about that.
I have done model distillation of XLM-RoBERTa into ALBERT-based models with multiple layer groups and for the tasks that I was working on (syntax) it works really well.
E.g. we have gone from a finetuned ~1000MiB XLM-R base model to a 74MiB ALBERT-based model with barely any loss in accuracy.
A better rule of thumb is miles because hours don't reflect the amount of energy put into your drive train.
Typical recommendations are every 500-750miles for a mountain bike. For the average mountain biker I think that would be more like every 125 hours at 6 mph. Mountain bike chains see much more abuse than road, and you can go further too.
You can be more scientific by using a chain ware tool to measure how much stretch there is.
You can't really make all things be equal and get increased efficiency and longetivity for most cases.
The trade-off are explicitly made to have reduced longevity to have better efficiency.
Take tires for example. A low rolling resistance lightweight XC tire will have significant advantages over a dual casing 2.6" DH tire. The XC tire will smoke a DH tire but not have better longevity on dirt. The XC tire will be slower on some DH tracks and the DH will be slower on XC. If you ride a DH tire on a road it will ironically wear down more quickly than a typical XC tire, and you can shred an XC tire in one day on technical DH.
Years ago I worked at a small, like 5-7 person startup in soma where we were surprised to find Rupert himself personally come by, later send some chiefs to do due diligence, and cut us a check. We worked with a few of his brands after that.
It didn't feel different than any other VC experience I've seen as an employee, whether that was A16z, Sequoia, etc. If it looks like a duck...
Blaine Washington is right along the Canadian border and next to many islands. The islands are sparsely inhabited and infrequently visited by humans it makes me wonder if maybe there is still a colony out there. It seems like a difficult area to do containment so glad they tracked these down.
It looks like in 2019 Vancouver island (which is huge) had a colony get eradicated, but otherwise haven't seen much else reported.
A common approach in rendering engines to convert screen space coordinates to objects is to render a second image with light and shadow disabled where the color uniquely maps to an id. You then can uniquely identify 24 bits worth of objects without needing to maintain a KD tree.
The most accurate description is probably the authors built a hearing aid from the 1960s with only $1 today [0]. Its worn around the neck, doesn't have any fitting, doesn't appear to have beamforming or feedback management, and low amplification.
It turns out these things are all important in clinical outcomes for listening comfort and intelligibility. I've built a modern hearing aid and done extensive patient testing - the details matter and it's not just electrical or algorithm problems but tough mechanical and UX challenges to get something more state of the art.
While I haven't worn many neck mounted devices, the Bose Headphones can get crazy feedback and the feedback path is similar. This would cause squealing discomfort for patients and everyone around them.
The other issue to contend with in his grandparents demographic is the perception of being deaf and being seen wearing such a big bulky item. Image consciousness is a large hurdle to overcome these
Great ingenuity but commercially it's hard to see it take traction
Both https://audatic.ai/ and https://whisper.ai/ are taking the ML approach for speech enhancement, and the large companies are starting to get on board. It's the future!
To make a product that Apple believes is significantly better than the competition they had to design a very intricate solution that includes:
* High end look and feel not similar to the bulk of their products with lots of textile webbing and memory foam ear cups. These get way more wear and tear than normal electronics being exposed to sweat, sunscreen, etc. On top of that these must be safe for long skin exposure and comfortable across many head shapes and sizes. * High quality magnets with custom speaker design for low THD and large frequency range. * 2 custom ASICs built for sound processing and low power bluetooth and 10 audio cores each. * 10 microphone array for ANC and wind noise cancelation. * Multiple accelerometers + gyros head tracking with spatial audio.
It you remove one of those components, I'd be surprised if Apple still shipped this. At least 40 people worked on designing and engineering this new headphone. There are not many companies in the world with the right kind of talent for this and Apple happens to be one of them.