Since I'm writing this from KubeCon in Barcelona, I thought it would be apropriate to start off with a relevant link. I couldn't find one because computers are boring now so here's one about the other kind of container and associated hundred-ton robots.
We talked about encoding neural nets into pneumatics, now how about encoding neural nets in free-space optical networks? Or (Ok, they don't actually encode a neural net in this one but math is math, right?) in styrofoam and microwaves.
Good, yes please build a neural net to wirehead monkeys, I'm sure that won't come back to bite us in the ass in any way.
A computer vision article completely unrelated to the subsequent link: Adversarial Patches.
Ceph lets you embed arbitrary C++ into the OSD processes to push computation down to where the data is. Computation like, for example, Postgres query processing.
Speaking of C++, you know what it desperately needed? A JIT compiler.
Did you know that NASA runs a competition to build mining robots for use on Mars? Or that it has been going for ten years? Or that the Alabama Astrobotics team has won every single year, most recently with a fully autonomous run? Next up: robotic tunnel rats.
From the AAAAAAAAAAAAA department: The prospect of robotic tentacle monsters is unsettling enough even before building soft, pneumatic, EMP-proof brains for them. Why would you then jam the "squick factor" knob to 11 and tell me you could just index match the whole thing and have it be invisible under water?! I wonder how well ML models 3D print into fluidics...
I feel like the lesson of "do not train your AI with content from Bad Places" has been learned once already but here we go again (h/t a reader!). (Someone train it on D&D rules lawyer threads and see how long it takes to build a BBEG named Kansto dedicated to optimally packing NPCs). Hilarity aside, I broadly agree that the results are pretty damn impressive, feel like young-brain instead of dumb-program, and justify some worry about script-kiddie usage.
The problem with not being a statistician is that I don't know enough statistics to determine if this statistical package is reasonable. Sounds neat though.
Now that Elon (call me!) is building a HEAVY METAL STARSHIP, the next obvious question is how hard is it to make steel on Mars? Seems doable, actually (via).
Pine64 announces just a bunch of stuff, including a nicer laptop and a network camera.
Based on my MD from the University of Wikipedia, I've always thought that attaching prosthetics to the bone seemed like the right long-term strategy. It appears Europe agrees with me.
You can use GPUs for much more important things than just being silly and going to Mars. For example, you can generate an infite variety of nonexistant people, whether they're real or um, less so.
I vaguely recall reading about Clovis and Tang a few years ago but they've graduated to RHEL now. I wonder how much work it would be to back the server side with a small HSM...