“Google AI refusing to tell the truth about Tiananmen Square. When is Congress going to wake up and realize these tech companies are totally compromised by China. They’re killing our kids while vomiting Communist propaganda”
Beware of wealthy people saying they wouldn't do it again. They are just validating their own lifestory: it was hard, incredibly hard. Implicit in that statement are approximately 8 billion other statements, where they believe they suffered more than than every other individual on Earth who makes less than them.
Exactly, would he really rather be another worker drone, perhaps desperately looking for a job, having been 'let go' by the latest round of owner-class imposed austerity, and perhaps facing looming unpayable rent bills etc?
This is doable. Many coastal wastewater systems already have large pipes that extend miles to sea (1), and that's a rounding error compared to the many more miles of pipe routing sewage to the plant (2, pg A-3)
Has anyone tried the same adversarial examples against many different DNNs? I would think these are fairly brittle attacks in reality and only effective with some amount of inside knowledge.
Yes. It is possible to generate one adversarial example that defats multiple machine learning models -- this is the transferability property.
Making examples that transfer between multiple models can affect "perceptibility" i.e. how much of change/delta/perturbation is required to make the example work.
But this is highly dependent on the model domain. Speech to text transferability is MUCH harder than image classification transferability, requiring significantly greater changes and decreased transfer accuracy.
I'm pretty sure there were some transferable attacks generated in a black box threat model. But I might be wrong on that and cba to search through arxiv right now.
Author here. Some of them are black box attacks (like the one where they get the training data out of the model) and it was done on Amazon cloud classifier which big companies regularly use. So, I wouldn’t say that these attacks are entirely impractical and purely a research endeavour.
> coupling, decoupling and light guiding elements to divert incident light to a concealed sensor
So, there's a camera in the dash looking up at the windshield and focusing where it expects to see a face, thereby using the windshield as a reflector? And maybe there's some additional etching and deposited films in the windshield to support the angles required?
And perhaps you can put cameras elsewhere, and similarly subtly modify the windshield or other glass to look at other things as well?
Wow. The original article was talking about doorbells and TV broadcast applications. That looks like the quality of the original moon landing broadcast.
I've lived with an EV for 3 years now, and take some legit road trips from SF: Utah, Vegas, LA, San Diego, Tahoe. I have also been all over the world and done many, many trips (far to many to count). I submit the optimum mix for most Americans right now is roughly 1 EV and 1 hybrid, solar roofing and a battery storage system. The average American has a spouse. Even if we ignore kids, we can roughly assume, unless those suburbs are way more empty than they appear, that both people have a car. Let one have an EV, and one take a hybrid. That way, they can cover the occasional very long drive in relatively remote areas.
Keep in mind, it may very well eventually switch, where gas stations are less common than high power EV chargers in the remote areas. Sort of a Dutch disease issue: once the EV chargers are the dominant market, the gas station market is likely to quickly fade until it's just diesel and finally all electric.
The Mad Max theorists worry that they won't have power for their electric vehicles in the event of an apocalypse. Friends, how long do you think refineries, pipelines, and oil freighters are going to stay going in the event of an apocalypse? Better to get good at rigging some salvaged solar panels, an inverter, and re-learn the old pass times, like dominos, dice, and cards.
Snow and ice. Deep snow, deep ice. Layers of ice, then snow, then ice, created within a couple hours. I think even if it slides off, that could cause some damage.
It's a holdover from the days of Morse code. Recall the first computers were for military problems, and the first output was teletype. Teletype was originally for military messaging and that had a long history of using all caps because they relied on manual transcription of Morse code (and other codes) over wire and radio. The all-caps policies were put in place to make sure the officers could consistently read what the operator had transcribed. Some of these date back to the 1850s. The Navy didn't actually do away with all-caps until 2013.
I think this makes more sense (or at least gets more directly to the point) than any of the Stack Overflow answers. Lowercase is easier to read in print as our minds learn the shapes of the words and can interpret whole words at a time rather than letter by letter. But Morse code was originally transcribed by hand, and it is easier for sloppily written lowercase letters to be mistaken for one another than the more distinctive uppercase letters, so it became a standard to write in uppercase. This tradition carried on to teletype and early terminals until both cases were supported.
As someone who had extremely poor handwriting as a child (it is still not great...), it makes a lot of sense to me that they'd land here.
Over the years, being regularly mocked/embarrassed/reprimanded over my handwriting and often forced to re-write assignments led me to develop a weird print ~hybrid casing that substituted a fair number of uppercase forms anywhere my lowercase forms caused trouble.
(This is mostly a fallback when someone can't read my cursive, or for official forms, package labels, etc. For the same reason I also adopted a smaller number of uppercase print capitals in my cursive.)
When it isn't socially appropriate to use ALLCAPS or come across as a sTRANgE pERsoN, I have to be fairly careful/attentive when writing in print to avoid dropping into mixed case.
(I'm not a monster; I'll scale these more like smallcaps when I write them.)
It's been a few decades, but I recall having to write in all caps in my high school drafting class. We were told the all caps and the letterforms we were told to use helped legibility when the drawings are stored on microfiche.
We were still taught that in high school in the late 90s. Probably some of the last. You are correct. Here's a good example of drafting from the 1960s: https://i.imgur.com/tXozAMy.png
It's not just all caps, but a flowing, easy-to-read style of all caps.
Hard to believe people used to draw all this stuff by hand. I was never very good at it.
The header image on the website shows slashed zeroes, but the font preview doesn't, which is unexpected. In fact when previewing in Google Fonts there's barely any visual distinction between uppercase letter O and zero. There's only a very slight difference in width which you actually have to look for. Punctuation isn't centered in their box and leaves so much tailing space the test text `5240N 15:01 .84 8,9` reads like `5240N 15: 01 . 84 8, 9`
These problems don't exist as much in the variable width font, where there is no tailing space in the punctuation and the uppercase Os are rounder.
There's also some weird pinching in the corners of N. 1, l and I are perfectly distinguishable though, so at least it's got that going for it.
The hollow zero vs slashed or dotted zero thing is often done in fonts with an extra variant; basically the font can contain both and will display what the application requests. So you should assume that the font really does have the slashed zero, but the font viewer you looked at it with didn't ask for it.
That's a good point. I added that rule (font-variant-numeric: slashed-zero;) to the font preview using the browser's devtools, and I saw no difference.
Also, for decades teletypes used 5-bit 'Baudot' codes. 2^5 = 32. (LTRS) A shift-key (FIGS) could be used for number digits and most punctuation. (Telegrams shown in old movies are almost always upper-case only.)
7-bit ASCII (2^7 = 128) didn't start to show up until 1963. (Starting the arguments.)
Funnier: the machines used gears inside that limited how fast it could type. (60wpm on the low-end ... Which also made it hard to 'read the mail' of commercial carriers that used 66-75wpm machines.)
This comes down to a combination of security and provenance. You have to protect the data, of course, but you also have to keep track of the use rights. Identifying all use rights upfront seems tricky, are the Creative Commons licenses sufficient?
This BNN, btw, is basically a playground for an advertising wunderkind
https://bnnbreaking.com/about
https://en.wikipedia.org/wiki/Gurbaksh_Chahal