I really liked this article until I went out and looked at how much Leica M2s cost. The project may only cost $100 in Raspberry Pi parts, but it looks like acquiring a Leica M2 can cost thousands of dollars.
You can always take the soviet rip-offs like the Zorki 4 [0]. Price is around $50 (at lest the last time I checked). I have one (it was my first cam gifted to me back in the 1980s by my father) and it still works very well. Though I nowadays tend to adapt the Jupiter 8 lens I got on it to my Nikon Z6. :)
Zorki cameras are Leica II/III(aka D2/D3) clones with LSM/L39 mounts, which don't take later M-mount lenses, though M series bodies(M3-M7) takes LSM via an adapter. Also they're not real Leica in case you care.
Indeed - that made me laugh. I've a single ZM lens left over after I sold my M9 and I'd really like to get it back to active duty. The price of ANY M-mount camer body is excruciating.
I really liked the project info regarding shutter speeds etc, and I'm considering trying this out with a Nikon FE2 I don't really like and picked up for £100 some years ago, but I suspect even these cameras have gone up in price. I might make another pinhole camera for the RPI sensor, that can be done with an old body cap and any camera-like box.
Speaking of Nikon, the Nikon F2, a fantastic, fully mechanical[1] SLR, is widely available on the used market for <$500 and compatible with nearly every full-frame Nikon SLR lens ever made with an aperture ring[2].
My current personal favorite "vintage" camera is an F2 with non-metering prism, type H full-field microprism viewing screen, and 50mm f/1.2 lens (still a current[3] Nikon product!), mostly because it has the clearest, brightest, largest viewfinder image I've ever seen on a 35mm camera.
[1] Other than flashes and optional electronic metering prisms.
[2] The only exceptions I'm aware of are the PC-E tilt/shift lenses, which have "soft" physical aperture rings that still require electronics, and therefore power supplied via the lens mount, to actually stop down the lens.
Also, my claim only fully applies to non-metering F2 viewfinders; metering prisms protrude a bit from the front of the camera body and interfere with the mounting and/or operation of certain lenses (including, I believe, all PC lenses).
I had an F2AS for many years, excellent viewfinder indeed, 0.8x, the same as non-HP F3, but quite a beast to carry around. The ME Super had even bigger magnification but probably not a 100% coverage screen.
I only sold the F2 motor drive last year, and that was ridiculous too, ten AA batteries for the MB1 battery pack. Cannot believe I used to take that thing to parties for kicks 20-30 years ago.
Yes, people do make mistakes; often. Airline pilots have long checklists of things they have to do before taking off, and more checklists for dealing with problems that inevitably crop up.
The Copyright Office has decreed that AI generated content cannot be Copyrighted. If these books have a copyright listing then they are a violation of the rules. I doubt if the "author" is at risk for legal repurcussions, beyond not being able to sue for copyright violation.
There is nuance in the statement that you're referring to. Works created by an AI are not copyrightable, but works created by humans who use AI tools are copyrightable.
Non-humans do not have copy rights from the things they create. However, humans have copy rights even when they use tools.
I have yet to see a legal argument that says the output of ChatGPT is free from derivative works claims. So you can't assert ownership of the output, but that doesn't mean someone else can't assert rights on the grounds of derivative works.
It is one of the reasons I don't see Copilot going anywhere. Any company that uses it can't assert ownership over their own IP, and if Copilot accidently reproduces a chunk of training data verbatim then someone else can.
I doubt many codebases are 100% generated by Copilot with no human input, and it’s not like having some Copilot snippets in your code is going to invalidate the copyright on the whole thing… I think that’s like being worried that writing parts of a book in a public domain font is going to invalidate the copyright on it
The copying verbatim thing is valid though, although they have mitigations against that
I remember reading this quite a while ago. My elaboration of this technique is to write the last sentence in my head, but only put the first part down on (paper) the end of the file. If I'm lucky, when I come back to it, I can read the first part of the last sentence, and memory will tell me how the sentence ends, I can just keep going.
Rarely need to print anything at all - everything is digital. Bit like I don't have a fax machine.
For the 2-3 times a year I do need it...there are multiple shops in <5 min walking distance with printers. And if that fails I can always print at work.
I have the Freestyle Libre system. The probe that "pokes through your skin" only happens once every two weeks, and my experience is that inserting a new Freestyle sensor hurts less than the average finger stick blood test. Before I got the Freestyle Libre I was doing finger sticks 4 or more times a day. With the Freestyle Libre I am checking my blood glucose an average of 20 times a day, and if I download the data from the Reader I can see glucose readings every 15 minutes, 24 hours a day.
I'm waiting to go back to them, a bit of a juggle to get insurance to pay until the current sensor or maybe transmitter runs out. I got angry at their customer service when I knocked a sensor off, so I tried the Dexacom g6, but their software is insanely bad. On top of that their sensors are only for 10 days which is much more of a scheduling hassle. And after all the pain of their icky software the killing point is they have an unsilenceable "6 hours to sensor expiration" alarm that can and will go off in the middle of the night, unless you actually power down the device or your phone. I hated their device and the phone connected anyways. The Freestyle had much better software with sophisticated graphs showing trends over long time periods to spot typical low and high time periods.
You can silence the expiration warnings in Settings > Alerts > Scheduled, disable Always Sound. You will still be alerted for critical lows and actual sensor failure. I know you moved to Freestyle but other Dexcom users might be reading.
Yeah, I think they're a huge advance for people who are already used to stabbing themselves. But they're still invasive in the sense that they have to keep the probe inside your body for 2 weeks. If Apple can nail the light-through-the-skin approach, the audience size jumps by 1-2 orders of magnitude.
I'm 72 years old and diabetic. There are significant limits on what I can do. My wife and I will do 30 minute walks 5 to 7 days a week. We go fast enough to get my heart rate up to the lower end of the aerobic zone. I also do strength training at the YMCA 3 or 4 days a week. I use the machines rather than the free weights. I worry about getting injured using free weights. I can't do many of the exercises other people have described here, but I do what I can. My weight is in the middle of the Normal BMI range, my diabetes is reasonably well controlled, and I think my strength is good for a 72 year old.
My recommendation would be to find an exercise that you enjoy doing, do it often enough to improve your health, and take your time; don't rush it. If you injure yourself it'll set you back to the beginning.
>> I have no problem imagining a security camera application needing to monitor quite a few video channels.
As a joke I sometimes tell people the automatic flushing toilets in public bathrooms work by having a little camera monitored by someone in a 3rd world country who remotely flushes as needed, while monitoring a whole lot of video feeds. They usually don't buy it, but will often acknowledge that our world is uncomfortably close to having stuff like become reality.
On the inference accelerator? IIUC, the RAM is just to hold the model and whatever state it needs during a particular inference operation. I'm not an expert on ML but AFAIK 16 GiB is plenty. I suppose it'd also need to hold onto reference frames for the video decoding, but at 1080p with e.g. YUV420 (12 bits per pixel), you can hold a lot of those in 16 GiB. edit: e.g., 4 references for each of the 96 streams would take ~1 GiB.
Even on the host, 16 GiB is fine for say an NVR. They don't need to keep a lot of state in RAM (or for that matter to do a lot of on-CPU computation either). I can run an 8-stream NVR on a Raspberry Pi 2 without on-NVR analytics. That's about its limit because the network and disk are on the same USB2 bus, but there's spare CPU and RAM.
Opening more warehouses would require hiring more people. I remember reading that turnover among Amazon warehouse workers is exceptionally high. Sooner or later Amazon will have to raise what they pay these workers, which will drive prices for products sold on Amazon to increase.
I think it’s more like they need to make the warehouse jobs easier. Trying to get peak productivity out of people burns them out quickly and they go through so many people they actually burn through the entire job applicant pool.
