TIL: Google Gemini is significantly better at reading the kinds of colourful graphs you see on server monitoring dashboards than are any of the other leading #AI models from #OpenAI and #Meta. Why yes, I do, in fact, resent that I had to learn this. Because #OpenSource tools don't make data visualizations #accessible to #screenreaders (even on the terminal) sometimes the things I #selfhost to get rid of big tech just mean I have to use big tech anyway to fix them when they explode. But never the less, Google is the only AI that doesn't tell me about the menus and other Windows I captured by mistake in my screenshot, doesn't make unhelpful recommendations, and actually provides the data without cluttering it up with Emoji. But I did appreciate metta telling me "This graph shows CPU use on the X axis, RAM use on the Y axis, network use on the Z axis, and time on the fake axis." I mean, open source UI is famously terrible...but I don't think it's that terrible...yet! Shhh, don't give the data visualization people any ideas!

Mikołaj Hołysz reshared this.

in reply to James

@wiresmith So I'm afraid I have to give you the dreaded "it depends" answer. If I'm trying to find a process/docker/whatever that's using more resources than expected, I probably want a table sortable by all of the various columns, with a combo box to set the window to examine ("average over the last minute/hour/day/week"). If I'm trying to determine what, exactly, is causing that resource use, I probably want some kind of real-time values reporting. Apple's Charts accessibility API does a good job of this: VoiceOver can actually play a beep that pans left/right for one axis, and pitches the beep up/down for the other. There are also tools (though only for Windows) that allow me to monitor the CPU or Ram use of a selected process on the machine, by just playing a constant beep, and changing the pitch up and down as the value changes. This way I can trigger various functions, and see when exactly the use spikes. It can also be useful to be able to set "expected resource use", and then get back a list of timestamps when the value was above or below what I set as the "expected" range. Then, at least, I can start matching up these timestamps with other things happening on the machine (chron jobs? A popular website post? Whatever.). I think the ultimate solution, here, is to get all of the people making these dashboards to use SVG and other standards a lot more, instead of just rendering png/jpg graphics. This way, the accessibility tools could be built as browser addons, rather than making every single tool that offers any kind of visualizations re-implement accessibility over and over again.