To some extend I agree, that if you are professional in some are, current LLMs may not solve the issues you are facing. But, while I consider myself as "expert" in one very specific area, and don't find huge help from ChatGPT in that domain, I don't live in a bubble. I have to sometime touch on completely orthogonal skills where I am not expert at all. And in these areas, I get huge benefit even though LLMs give me "average" responses. It is still better than my own, below average judgements. That is very strong point.
Yes, that is a very strong point. And we all have differing levels of expertise in different areas. We are all pulled up to some extent.
The article, though, speaks about the context of work, research, and industry - the professional context. In strictly that context, if you are "the best" at your day job, perhaps you won't benefit much there. By saying that "the best" benefit more than "the rest", The Economist is completely at odds with reality.
I purchased Bose QC-35 II as well. I don't have chemical burns, but after a while of wearing my ears become really sensitive. I feel strange pain while wearing them. This applies only when the noise cancelation is ON. When it is off the pain immediately disappears. My explanation is that these phones actually generate inverse signal to the noise, but due to various factors such as delay, the cancellation is not perfect and what I am actually getting into the ear is the residual high frequency noise, which may in fact be quite dangerous. Anybody else experienced this? There might be some study about it ...
> but due to various factors such as delay, the cancellation is not perfect and what I am actually getting into the ear is the residual high frequency noise
If you model the noise canceling as a superposition of the input signal with the generated signal, where the generated signal is specifically designed to account for the processing delay and hence only targets lower-frequency noise, then wouldn't 2 things be true:
1) Any residual from imperfect cancellation would only be at the lower frequencies (because the generated signal itself contains no high frequencies, and we a linear superposition). Of course, the original high frequency content from the input that was never cancelled will still be present but this isn't any worse.
2) Even in the worst case that we completely mispredict, we will only double the sound intensity which is a 3db increase. This seems relatively safe?
That said, while it seems physically safe it's possible that ANC still wreaks havoc with the brain's audio processing. Maybe the brain relies on the existence of the low frequency content as a sort of gain control mechanism or something (there are anecdotal reports by some users of increased tinnitus with long-term ANC use, but it's also possible that it only increased their awareness of it). I've also read that some people are sensitive to this lack of low-frequency noise since the brain interprets it as a pressure differential.
My background is in image processing and I can see quite clearly that the high frequency noise happens when you subtract two images, one of which is shifted by half-pixel for instance. What you are left with is the edges (high frequency) of the image. The same applies to 1D in my opinion. For this reason I am not convinced (1) is actually true. The argument about lack of low frequency is interesting, who knows what is happening there indeed.
Images processing and traditional 1D signal processing are not directly comparable, especially when dealing with analogue signals. You can’t have a true “hard edge” in an analogue 1D signal, because it not physically possible to record or reproduce, so you’re always dealing with sinusoids.
If you take a sinusoids, invert it, slightly offset it, and combine them. You get a smaller signal always, unless you’re delay is larger then 90°.
I would recommend playing with some audio signals in an online simulator and see what you get, you realise that your 2D intuition does not apply well to analogue 1D signals. The strict digital nature of image processing done on a computer creates the possibility of results not easily possible when working on analogue signal. After all you can one pixel on a screen at max brightness, and it’s neighbour completely of, but it’s impossible to recreate a similar hard edge with an audio signal because it would require the speaker to be capable of infinite speed and acceleration.
You're simply wrong, here. Take a single sinusoid and do that, sure. Take a large number of sinusoids of different frequencies, as all music and public noise is, and invert/offset, and you get everything from double amplitude to zero amplitude. None of these simplistic mental models of noise cancellation will help.
Sure, but show me how you get significant frequency shifting, or injection of frequency components greater than found in the two source signals. That’s the topic of discussion, not amplitude changes.
There's no mention of frequency shifting or anything like it in the parent post. For amplification of certain frequencies in a signal via simplistic noise cancellation, just take a combination of a sine wave at 10kHz and a sine wave at 1kHz, offset it by 0.00005s (call it signal processing delay) and subtract it from the original. The 1kHz signal is basically (noise-)cancelled and the 10kHz signal is doubled in strength.
Not that any of this is relevant to the original problem.
> the cancellation is not perfect and what I am actually getting into the ear is the residual high frequency noise, which may in fact be quite dangerous.
From GP:
> I can see quite clearly that the high frequency noise happens when you subtract two images, one of which is shifted by half-pixel for instance. What you are left with is the edges (high frequency) of the image.
Show me the part where the word “amplitude” appears in any parent post.
Additionally as mentioned in parent posts, noise cancelling headphone run a low pass filter over the input to the ANC system, specifically because achieving good alignment between your ANC signal and original signal is basically impossible at wave lengths shorter as you can’t know the exactly which direction the original signal came from (direct perpendicular to the head, or at a close tangential angle), and thus can’t compensate for the offset needed to ensure the two signals arrive at eardrum at the right time.
Ah, you're assuming that everyone agrees that ANC always uses an early low-pass filter, and hence there'd be no high frequencies in the noise-cancellation signal. Makes sense.
However, I don't agree that ANC always uses a low-pass filter, and it seems from kalal's followup that they are also talking about the using the full original signal. So the two of us were not talking about introducing high frequencies but about somehow enhancing the high frequencies already present, and that's what the figures I gave above are for. So we've been talking across each other. I apologise for my part in that.
("Amplitude" was a simple technical term to replace woolly terms that were being used, just as you are the first in the parent-chain to say "low pass filter".)
> high frequency noise happens when you subtract two images, one of which is shifted by half-pixel
> What you are left with is the edges (high frequency)
So you subtract a slightly phase shifted high frequency signal, you're left with a high frequency signal that may be amplified at the edges depending on your phase shift. Nothing surprising here?
The question is can you create a high frequency residual by subtracting a lowpass filtered (gaussian blur?) image? I don't think so. You're just left with whatever high frequencies you had but you aren't creating any new ones.
Yes, that would be correct, under the assumption of a lowpass filter. As I've just mentioned elsewhere in the thread, though, I don't think kalal accepted that there was a lowpass filter involved, and even stated such, indirectly, and without that it's easy to have a delayed subtraction that amplifies higher frequencies, as in the image example kalal gave.
Active noise canceling makes some people feel like there's a pressure difference, like when taking off in a plane. It's because the headphones block low-frequency noises better than high-frequency ones, and the brain interprets a proportional lack of low-frequency noise as a sign of pressure difference across the eardrums.
It's worth noting that you're applying an overly simplistic model of what noise cancellation (at the Bose/Sony level) actually entails. There's a lot of signal processing involved and it's exceedingly unlikely that the headphones are significantly increasing the amount of high frequency noise reaching your ears. (Some noise cancelling headphones even have a microphone on the inside, to measure what your ears hear.)
There are lots of other sources of pain, fatigue and discomfort, though. Essentially, since the noise cancellation isn't perfect, you can hear some subset of the sound, varying across the frequency spectrum. This can cause a feeling of being underwater, of having blocked ears or simply of having to listen more closely, all of which might cause ongoing muscle tension of different muscles.
On the good news side, a lot of people who find discomfort with one brand of noise-cancelling can find other brands fine, so it might be worth trying some other brands. (For Bose, the main similar-quality competitor is Sony.)
Finally, in the simplistic model, there does not have to be a delay. Sound only travels at 340 metres per second, so for every 10mm distance between the microphone and the emitter, one can have around 34 microseconds of processing time.
I don't feel pain, but yes, noice cancellation on those Boses can have some weird effects in quiet environments. Therefor I only activate the noise cancellation on louder environments and then it really helps me concentrate.
But the noise cancellation is subtly audible. I'm not a sound engineer, but it feels a little bit like sound compression being too high.
I actually never cared for the noise cancellation on the QC-35 II. I did use it on planes though. I liked the feature a bit better on the earlier wired model. I think Sony has a comparable product where it has a finer grained toggle of the inverse wave.
I am not expert, but if you don't use noise cancellation you would probably do better with standard headphones where you pay for the sound quality instead.
The QC-35 II's were great for concalls. You can connect multiple devices at once via bluetooth and the mic quality was quite good. For music, I agree there's lots of better options. I've now using an external microphone and some wired DT-770's, however it's not a great portable solution.
I can confirm that I had a similar feeling after wearing the the AirPods Pro for just about an hour. I tested them for a few days and returned them in the end.
However, my Sennheiser over-ear ANC headphones (MB660) have a much better ANC effect compared to the AirPods but the strain on the ears is a lot less intense. I would expect, that the Bose QC-35 II are much closer to the MB660 than to the AirPods Pro.
I am honestly shocked by that! When you google authors name, you get bunch of anime faces + GAN research papers. So clearly, this is not a joke, but his personal interest/fetish developed after long hours behind his PC. I believe he needs medical treatment and not propose "startup ideas"!
I'm not sure if this is sarcasm or not, it's hard to tell over the internet. But if it's not, out of curiosity, how did you make the connection between anime and child pornography?
To a colorblind person (me included), this discussion is absolutely irrelevant. Why not to focus on edges instead?
A black and white picture would solve all such nonsense.
This is important point indeed. Here is what helps myself, when thinking about pixels.
- pixels are point samples in 2D space
- their position is exact
- position of of top-left point is (0,0)
- position of bottom right is (cols-1,rows-1)
This way all math work (subsampling, affine or perspective warps, lens distortion, or even warping between image and any layer neural network). Failure to do that, will cause subtle issues that will degrade your performance. These will become quite important when working
with pixel accurate methods (3d object tracking, object detection in tiny resolutions and 1-1 mapping between OpenGL and Neural Network). So I have to agree, pixels are not tiny squares, they are dots in 2d space.
I was wondering, how much different the simulation would be from stock models from Turbosquid: https://www.turbosquid.com/3d-models/formula-1-season-2020-3...
The article mentions there were some problematic areas in the extracted model. So I am quite skeptical that the model is indeed more accurate than what you can easily get from legal sources.
Organizations where the big $ is on intel property and R&D wouldn’t be dumb enough to give the actual schematics to some random intern building an app. There are some serious security controls and mNDA around that data, for all internal parties, external parties and suppliers. Think about it, they must collaborate with dozens if not hundreds of parties to get to a finished products, and some amount of information must be circulating to achieve the purpose of their relationships. If those were to leak - they’d leak to the competitors first and it’d be a disaster for the entirety of the upcoming year(s) as the improvements tied to their new designs would now be considered available on all cars. It doesn’t take two dummies to think about that, soooo.. to avoid losing hundreds of millions, what do you do as an organization? Data classification, risk assessments, tight security controls, procedures, and finally mNDAs for the instances when data do have to be shared.
Furthermore, as it was pointed out on Reddit, the model appears to be made of parts that are valid for 2020 regulations, some parts for 2021.. in other words, it’s just scrap.
reply