These techniques are used by DeepSeek, and work well with the commodity (NVIDIA) GPU's they use.
Google designs their entire AI stack from the custom silicon up. So they have different optimization approaches.
(Though Gemma does use MTP)
It's been reported Iran is deploying IR missles along the common ingress and egress paths the U.S. has been using for the past month. So target acquisition could be MK I eyeball
The Carl Gustav 8.4cm recoilless rifle now be procured by the Marines has a big back-blast through the typical venturi used by these kinds of weapons. In training, (Sweden at least) they are limited to the number of rounds they can fire. I do not know if this for merely hearing loss, or if there are concussive effects.
Also, while "small arms" do not typically have concussive effects, I have stood behind a.50 Barrett rifle discharge and there was a noticable shock wave in the air. I didn't feel it in my head, but my intestines were not happy. If you were in the front hemisphere of the muzzle blast it would be far worse.
Muzzle brakes on rifles usually make the blast wave propagate more to the side. A high caliber rifle with a muzzle brake could easily hit 160db. Maybe a .50 cal could go higher. So now you are even punishing your friends, not just folks downrange.
Really good article. Interesting to see how even Google is scrambling to keep TPU architectures up to date enough to be relevant for the latest model types.
I made a wood plate VESA adapter recently. I used a paper template for the hole spaceing, and a nail to mark the holes so the drill bit didn't lose centering when starting the holes. To make the template, I taped a piece of paper over the VESA plate and rubbed the side of a pencil lightly over paper on top of the holes. Leaves nice circular marks where the holes are.