tl;dr: Berkeley researchers trained models on a possibly-accidentally-leaked Chinese dataset -- images of US naval ships, suspiciously labeled with bounding boxes around their radar systems.
They're making the point that training models like this can give some intelligence towards another country's ML capabilities (here, China). In my opinion, a cool work that has way more "contribution" than your average AI conference paper.
(And full disclosure: I contributed a bit of feedback on their writing.)
That's really strange and incredibly frustrating - but slightly less so if it's consistent with all of the bars (including their own).
I take issue with their choice of bar ordering - they placed the lowest-performing model directly next to theirs to make the gap as visible as possible, and shoved the second-best model (Grok-1) as far from theirs as possible. Seems intentional to me. The more marketing tricks you pile up in a dataviz, the less trust I place in your product for sure.
Is the fall detection alt-text actually that good here? There were more elements in that image, which I assume people who use screen readers or are reading the alt-text for broken images would want to know.
The "EMERGENCY SOS" slider and "I'm OK" button give more examples of the "straightforward and direct" language that the article text references. I also learned something about that feature itself (besides that it existed in the first place) - Apple's design choices to make the "SOS" a slider, followed by a larger/easier to press button for "I'm OK". Even though it wasn't related to the point of the article, it was information that I wouldn't have learned had I just read that alt-text.
Is this part of accessibility guidelines for alt-text? Shouldn't they convey the same information, whether it's in image or text form, even if it's not directly relevant to the point of an article?
I can also imagine that people have different preferences - maybe some want all the information like I mentioned, whereas others don't want to distract from the point of what they're reading. I wonder which way the alt-text guidelines lean in practice.
I've tried, but it only works well if it's the beginning of a project. Later on, it gets annoying to sync back and forth between copies as others work on it, since I like to see what I'm writing in the context around it.
I'd like to try this, but it's tougher when you're collaborating on a shared document.
For example, I collaborate on papers that always start out as a shared Overleaf doc. When I'm starting a new section, I just want to toss some ideas down on the page, knowing that the prose is terrible. I can really enjoy this phase if I know it's free of judgment. However, it's so much harder when I know that someone might open the doc in an hour and judge my writing before I get to edit it - or worse, if I see that they've opened the page while I'm in the middle of it, or god forbid see their cursor click around near where I'm typing. It's a lot harder to concentrate until they leave or start editing somewhere else. I end up taking a while to write each paragraph, planning & editing as I go.
From a quick search, there are now sites like [0] where you can seemingly buy road signs which adhere to MUTCD (I can only imagine it was much more difficult for Ankrom in 2001). From the customer reviews, it sounds like many are placed within personal property, such as indoor decorations or arrows directing delivery drivers.
However, I wonder how much trouble people get in practice for putting up signs on less-trafficked public roads (e.g., signs like "Slow, children at play" or pedestrian crossings). I'd like to think that a well-intentioned sign on, say, a quiet suburban street would be allowed to exist.
What sorts of doctors can dictate their own schedules, work minimal hours, or take months of vacation each year? If they exist, surely they're firmly in the minority.
They're making the point that training models like this can give some intelligence towards another country's ML capabilities (here, China). In my opinion, a cool work that has way more "contribution" than your average AI conference paper.
(And full disclosure: I contributed a bit of feedback on their writing.)