I have been curious if targeting a virtual machine such as qemu would make driver implementation easier; only one disk interface, one network card, one gfx device, etc. since your just accessing what the vm gives you. Does anyone have experience with this?
I don’t personally but I’ve been following [https://www.redox-os.org/](Redox OS) and that seems to be their process for that reason.
(Note: markdown links have the text in the square brackets, and the link in the parentheses)
I fucking hate reCaptcha, partly because the problems seem to be getting harder over time. Sometimes I literally can’t spot the cars in all the tiles.
It’s also very effective at keeping Tor out. ReCATPCHA will, more often than not, refuse to even serve a CAPTCHA (or serve an unsolveable one) to Tor users. Then remember that a lot of websites are behind CloudFlare and CloudFlare uses ReCAPTCHA to check users.
For the Cloudflare issue you can install Cloudflare’s Privacy Pass extension that maintains anonymity, but still greatly reduces or removes the amount of reCaptchas Cloudflare shows you if you’re coming from an IP with bad reputation, such as a lot of the Tor exit nodes.
(Disclaimer: I work at Cloudflare but in an unrelated department)
Luckily, CloudFlare makes it easy for site owners to whitelist Tor so Tor users don’t get checked.
Realistically, how many site owners do that, though?
I don’t hate it because it’s hard. I hate it because I think Google lost its moral compass. So, the last thing that I want to do is to be a free annotator for their ML efforts. Unfortunately, I have to be a free annotator anyway, because some non-Google sites use reCaptcha.
Indeed, also annoying is you have to guess at what the stupid thing is trying to indicate as “cars”. Is it a full image of the car or not? Does the “car” span multiple tiles? Is it obscured in one tile and not in another? Which of those “count” if so? Should I include all the tiles if say the front bumper is in one tile or not? (my experiments have indicated not).
Or the store fronts, some don’t have any signage, they could be store fronts, or not, literally unknowable by a human or an AI with that limited of information.
I’m sick of being used as a training set for AI data, this is even more annoying than trying to guess if the text in question was using Fraktur and the ligature in question is what google thinks is an f, or an s. I love getting told I’m wrong by a majority of people not being able to read Fraktur and distinguish an f from an s from say an italic i or l. Now I get to be told I can’t distinguish a “car” by an image training algorithm.
At some point, only machines will be able to spot the cars.