2,142 karma · joined June 28, 2022
Not all systems upgrade every few months. A major question is when we reach step-improvements in performance warranting a re-eval, redesign of prompts, etc.
There's a small bleeding edge, and a much larger number of followers.
Most modern cameras can stream video to a computer through a proprietary protocol. These are implemented under Linux in gphoto2, and in other OSes, through some proprietary tool. During the great webcam shortage of covid, many companies made special, flaky Windows utilities to allow those to be used for web conferencing. Very few can natively as a USB Video Class (UVC) device. This is Canon's version:
https://www.dpreview.com/news/4796043082/canon-s-new-softwar...
Now, for Canon, it's a monthly subscription:
https://www.usa.canon.com/cameras/eos-webcam-utility
As a footnote: The general rule-of-thumb is about f/11 is where you start to notice diffraction limiting sharpness on full frame. That's a rule-of-thumb, and you're welcome to not step down below f/8, but calling f/9.5 "abysmal" is more than a little over-the-top. But no, a phone will not compare to a full frame with a $2000 f/1.2 lens. But it's quite competitive with a kit lens.
All this went away. I felt a loss of joy and nostalgia for it. It was bitter.
Not bad, but bitter.
People are leaving off which lens. In my experience, for low-light:
Large sensor + kit (zoom) lens < Pixel Pro < Large sensor + f/1.4 prime
It's not apples-to-apples, since my phone has no optical zoom in the lens (although it somewhat makes up for it by having wide/normal/tele fixed lenses). But shooting with the main lens, it definitely beats a large sensor for low-light with a kit lens.
I think the key difference is intelligent multiframe denoising algorithms on the phone. It, in effect, shoots a video and combines.
Sales numbers tell a different story.
> Artistically speaking, phones can't do what even a low end slr/mirrorless can do, its just that phones are good enough for the low-effort content 95% of people are interested in producing.
This is not correct.
A Pixel Pro has a 50 MP, f/1.7, 1/1.31" sensor. This is equivalent to f/4.6 in u43, f/6.6 in APS, and f/9.5 in FF.
This is slightly slower than a kit lens on paper, but this is more than made up for by more advanced sensor technology, and especially the ability to do things like fast sensor readout, which can read out many frames and combine exposures.
Side-by-side, shooting with a phone and a Panasonic u43 camera with a kit lens, I was getting perfectly good photos with the phone, and useless photos with the u43.
> I don't know about the manufacturing or drone stuff, but for video conferencing and remote education, the point of the video really isn't image quality or "art" but just good enough picture to not get in the way of the real purpose of the interaction, so a whole camera kit is just added complexity/annoyance for no benefit.
It depends on the context. People buy $100k Cisco remote conference rooms for a reason.
I've definitely spent >$10k on equipment in remote presentation / education contexts myself, and know many other people who have done likewise.
You should, at some point, figure out what popular education Youtubers, twitch streamers, etc. spend :) But there are similar contexts in scalable education, various kinds of sales, etc.
One of the core issues -- in context I've worked in -- is that reliability is king. I don't want interruptions. I'm happy to have three cameras feeding into OBS and a set of fixed setups, and I've even done custom plug-ins, but something like a mirrorless adds layers of complexity which can lead to bugs:
- Mirrorless
-> HDMI out
-> Elgato
-> USB
-> OBS
-> Virtual camera
A direct USB connection would remove a cable and an adapter.
There is a long list of issues like this which have prevented ecosystems from forming around cameras, in the way they have around Android or iOS. It's like the proprietary phones predating the iPhone.
The irony is that phones are gradually passing dedicated cameras in increasing numbers of respects as cameras are now in a death spiral. Low volumes means less R&D. Less R&D and no ecosystem means low volumes. It also all translates into high prices.
The time to do this was about a decade ago. Apps, open formats, open USB protocols, open wifi / bluetooth protocols, and semi-open firmware (with a few proprietary blobs for color processing, likely) would have led things down a very different trajectory.
Sony is still selling cameras from 2018:
https://electronics.sony.com/imaging/interchangeable-lens-ca...
The price new fell by just 10% over the 7 years ($2000 -> $1800).
And in a lot of conditions, my Android phone takes better photos, by virtue of more advanced technology.
I have tens of thousands of dollars of camera equipment -- mostly more than a decade old -- and there just haven't been advancements warranting an upgrade. A modern camera will be maybe 30% better than a 2012-era one in terms of image quality, and otherwise, will have slightly more megapixels, somewhat better autofocus, and obviously be much smaller by the loss of a mirror. Video improved too.
The quote of the day is: "I wish it weren’t like this, but ultimately, it’s mostly fine. At least, for now. As long as the camera brands continue to work closely with companies like Adobe, we can likely trudge along just fine with this status quo."
No. We can't. The market has imploded. The roof is literally falling in and everyone says things are "fine."
Does any know how much volume there would be if cameras could be used in manufacturing processes for machine vision, on robots / drones, in self-driving cars, on building for security, as webcams for video conferencing, for remote education, and everywhere else imaging is exploding?
No. No one does, because they were never given the chance.
And they've actually gone up in value, by quite a bit. $90k-$190k used, $25k new. Which is curious, since a decade ago, they were quite cheap.
1) CO2 levels have risen from under 300PPM in 1860 to over 400PPM right now -- by around 150PPM -- with a rise of about 25PPM per decade for the past four decades.
The difference in CO2 levels in my bedroom with windows open and closed is a couple hundred PPM (500-800ppm range in my bedroom, with windows open and closed, respectively). I can definitely feel a difference in performance if I don't let in fresh air. It's more than climate change (300ppm versus 150ppm range), but not big-O more, and climate change is on-track to get there in another few decades. Conference rooms might be over 1000ppm, but it still big-O similar.
2) CO2 levels are measured in parts-per-MILLION. That argument simply doesn't make sense. The atmosphere is 21% oxygen. Crowding out oxygen is simply not an issue. Critically, from personal experience, if I have some dry ice in a room, I generally don't suffer.
People run into problems when CO2 levels reach a out 5000 ppm over many hours. Even the most dense conference rooms don't hit that.
Many of those gasses do impact cognitive performance. It's not obvious to me why CO2 would, but if CO2 is going up, so is everything else we breathe out. CO2 where I am is somewhere in the ≈400ppm-1000ppm range -- 0.04% or 0.1% -- and it's pretty inert. I'm not sure what harm it does.
If it does to harm, rising CO2 levels should be much more concerning than "just" climate change.
But I suspect it's other gasses.
Until AMD fixes that bug, NVidia will continue to outflank them at every point.
Ballmer had it right: "Developers, developers, developers."
However, you get major successes by taking risks. Evaluating people and businesses, I do so based on successes and not on failures. The best people (and businesses) I know failed more times than successful ones. Cybertruck was a good try, and it takes a few like that to be successful.
Of the problems I have with Tesla, the Cybertruck is not one of them.
Cybertruck owners, on the other hand, you might as well but a dunce cap and a t-shirt which says "I'm an idiot with too much money."
That's a deep problem. Checks-and-balances are hard, and Google had none. Ergo, it went from good to evil.
For example, Larry and Sergei didn't seem like psychopaths. Neither did Omar Bose. Many other founders aren't.
Ditto for most processes with extreme selective pressures (e.g. most senator, PoTUS, many elite academics, elite law firm partner, etc.). People win by optimizing to the process. People with ethics, or who don't cheat, don't win. There are exceptions, but they're in the minority. For example, if you are an extreme genius or stumble on a major breakthrough, in science, and don't have it stolen by a psychopath.
You got downvoted, so I'll provide two references:
* Power by Pfeffer
* Dictator's Handbook by Bruce Bueno de Mesquita
Very different takes, but same general conclusion. Both are respected academics at Stanford and NYU, respectively, with de Mesquita being far more respected, and an excellent read. Power is a bit less so
What's been amazing to me is how skin-deep the stated values across the tech industry were. Virtually every executive is behaving in ways they would fire their employees for just a few months ago. The executives have no moral compass as human beings. I believed this to be the case before, since the people I've interacted with who rose that high did so by being psychopaths, but what's amazing is how on display it is.
I don't mind the before or the after so much as the very, very transparent about-face.
When I did the math with my life savings, 0.5% was about the breakpoint where things become significant. $200k would be more significant, so I suspect when I did the math properly, I was expecting to retire in less than 50 years. In 50 years, I'll likely be dead.
However, if the fee structure is 0.07%, that's $70/year / 100k invested. Even if it's 0.44%, you're talking about $440.
The fees on most funds are small enough now to not matter much. It's worth shopping for lower-fee funds, but the more you go below 0.5%, the less it matters. If I save $500 per year for 50 years, that's $25k+interest, which is kind of the breakpoint of where it has practical impact on e.g. when I can retire.
Intel 6950X is unsupported. That's 10 cores at 3.5GHz. It's about 10x faster multicore and 2x faster single-core than a modern, supported Intel Celeron N4500.
Moore-style scaling where chips are faster (rather than just denser) ran out a while back. Expensive computers from a decade back beat cheap computers of today. A modern 20-core PC is perhaps 1/10th the CPU cost as compared to when the 6950x came out, but not a lot faster.
If that 6950x was maxed out to 128GB RAM and set up with a decent GPU, it can be a very good system.
That's among the oldest processors not supported, but many computers with somewhat newer CPUs don't work either due to motherboard / BIOS issues.
That's a large n. I used them commercially.
However, it's also pretty old data (probably about a decade back).
Consumer Reports reported on it as well.
Suggestion: It would be nice to include a price list on the article.
This project is impressive as heck, but aside from being intellectually out-of-reach for most kids, it would be financially challenging as well. Last I looked, CNC aluminum blocks were well out of the reach of 99.9% of kids (but that was decades ago; perhaps prices went down).
For people wanting to follow in those footsteps, it'd be nice to know which things cost $5, which $50, $500, or $5000. Just that kind of intuition is helpful.
Honestly, the problem isn't just which devices, but even more so, this (from the page, not your comment):
> No guarantees of future support but we will try hard to add support.
During the Great GPU Shortage, I bought an AMD RX5xx card for ML work. It was explicitly advertised to work with ROCm. Within a couple of months, AMD dropped ROCm support. EOLing an actively-sold product from being used for an advertised purpose within the warranty period was, if I understand consumer protection laws in my state correctly, fraud. There was no support from either the card vendor (MSI). No support from AMD. No support from the reseller. Short of small claims, which was not worth it, there was no recourse.
This is on a long list of issues AMD needs to sort out to be a credible player in this space:
* Those are the kinds of experiences which cause people to drop a vendor and not look back. AMD needs to either support cards forever, or at the very least, have an advertised expiration date (like Chromebooks and Android phones).
* Broad support is helpful from a consumer perspective from the simply pragmatic point of view that only a tiny fraction of the population has the time to read online forums, footnotes, or fine print. People should be able to buy a card on Amazon, at Best Buy, and Microcenter, and expect things to Just Work.
* Being able to plan is essential for enterprise use. I can't build a system around AMD if AMD might stop supporting their platform on 0 days notice, and the next day, there might be a security exploit which requires a version bump.
I'm hoping Intel gets their act together here, since NVidia needs a credible competitor. I've given up on AMD.
That wasn't actually my statement. My statement was that it's a better camera.
My Android will actually take dramatically better photos than a u43 camera with kit lens, especially in low-light. It's not even in the same ballpark. My full frame with a Zeiss f/1.4 lens will take better photos than my cell phone. My cell phone also won't reasonably zoom (although it does have a telephoto of sorts and a wide angle, it can't compete with a 200mm let alone a 500mm on my full frame).
However, holistically, it's a better camera. If I take a photo with my cell phone, I can be editing it on my computer in a few minutes. If I'd like a different display -- for example for framing -- I switch apps. That's not to mention panoramas, photospheres, and other computational photography features. Computational stacking for low light can give pretty impressive results too. The list goes on and on. This paragraph could be an essay.
With the exception of sensor size, zoom, and clicky buttons, my cell phone does so much more. Those are important, mind you, but something like a 50 MP, f/1.7, 1/1.31" sensor on a Pixel 9 Pro is about equivalent to an APS camera with a kit lens for equivalent aperture. Under most conditions, the photo quality from a good cell phone is indistinguishable from a big camera. So all the other stuff does become more important.
I did pick a cell phone for its photography features. It's not a random Android phone.
If these opened up, at least to the level iPhone did in 2007, they'd have an ecosystem as people still used them. As-is, for most purposes, my Android phone is a better camera than my full frame interchangeable lens camera.
See:
Bruner / Spiral Curriculum.
Ebbinghaus / Spacing effect
Hattie / Deep-surface-transfer learning
Chunking ("How People Learn" has a good copy on this)
Etc.
The way you do this is you take a course, and then you take more courses. After a few years, it all connects and makes sense. The first course, I find, is often best short, simplified, and applied. Once you get through that, you can go deeper.
Different angles are nice too. For linear algebra:
- Quantum computing
- Statistics and probability
- Machine learning
- Control theory
- Image processing
- Abstract algebra / groups / etc.
- Computer graphics
All come to mind.
On a mile-high level, this course seems ideal for a first pass. On a detailed level, I'm confused by some licensing issues.
An older 3060 12GB is also a better option than the B580. It runs around $280, and has much better compatibility (and, likely, better performance).
What I'd love to see on all of these are specs on idle power. I don't mind the 5090 approaching a gigawatt peak, but I want to know what it's doing the rest of the time sitting under my desk when I just have a few windows open and am typing a document.
I don't understand AI TOPS. Per NVidia:
* 5090: 3352
* 4090: 1321
* 3090 Ti: 320
* 3090: 285
* 3060 Ti: 130
* 2080 Ti: 114
source: https://www.nvidia.com/en-us/geforce/graphics-cards/50-serie...
What's increased tenfold since Ampere? Can someone explain this to me and how it impacts real-world performance?
Autism is a disorder, diagnosed in DSM V. Very few people in your life are likely to be autistic. As with many such things, it's a spectrum. Unless those cross a pathological threshold, it's not autism. Few SWEs are over that threshold.
We used to call people "nerdy," "quirky," or other things (and it was okay to be that). That's in the natural spread of human personalities. If you're 5% of the way towards dyslexia, autism, obsessive-compulsive disorder, or whatever else, it's in the range of human personalities, not disorders.
It's helpful to understand how different people work. Most SWEs and introverts share some characteristics with people with autism, and reading about autism can help understand oneself by interpolation. However, the label is not fair either people with actual autism, or to people without autism who believe they have it.
Footnote: My experience is that extrovert versus introvert is mostly a function of whether you're invigorated by interacting with people or tired by it, and especially by people you don't know well. Yes, social skills are helpful, a skill one can develop, and important to one's career, but so are math or medicine. A good strategy for an introvert is to partner with an extrovert.
I'd totally do that if I got everything shipped to me, and knew I wasn't forgetting something.