buildbot

44 ms·

721.
by buildbot 3y ago
For the ultimate nerd snipe, how far would the light from that explosion be visible?
722.
by buildbot 3y ago
Yeah same haha - for a slightly more relevant number, if every nuclear weapon yielded 50MT of TNT, the asteroid impact is approx 63 Billion times the entire worlds nuclear arsenal in terms of power.
723.
by buildbot 3y ago
A relativistic asteroid would not be a good time for earth. Apparently according to gpt4 (so big grain of salt), a 1km asteroid going 0.9c is roughly 4.37×10^16 megatons of TNT worth of energy.
724.
by buildbot 3y ago
Seems interesting - if I understand, this lets you get Verilog from a frontend you build. Can this be used to implement say a backend for pytorch? How does one link the generated verilog to software that can drive the design (in sim or actu
725.
by buildbot 3y ago
I have this motherboard - a big downside is many of the PCIE slots will overhang into the RAM if used for a GPU. I can't use two channels in my current ML machine because of this, and I have single slot 4090s.
726.
by buildbot 3y ago
Lots of people really underestimate the impact of that mental state and the activation energy it creates towards doing experiments - having some local compute is essential!
727.
The Layoff(xeiaso.net)
1 points by buildbot 3y ago|0 comments
728.
by buildbot 3y ago
Yeah this page is full of straight up lies? “ Its performance in every regard is almost unreal (up to 284 times faster than x86).” Like, there are at least 3 things wrong with that statement!
729.
by buildbot 3y ago
You’d be surprised at how cheap high end networking that outperforms PCIE4 x4 is - 100Gb omni-path nics are running for 20$ on ebay! And those will saturate PCIE3 x16. Though of course with multiple boards/ram/cpu it gets complica
730.
by buildbot 3y ago
Interesting, I would expect the impact to be noticeable at 4x! and yeah it heavily depends on model, sharding method, model vs data parallel. I’m hitting the peak bandwidth due to a very wide, shallow model that is split between each GPU mo
731.
by buildbot 3y ago
"I started experimenting with Nvidia's RTX 4090s. I bought a bunch of them and put them into a mining rack and just ran some tests. I quickly figured out that is not the way to go," Well, I hope they were at least smart enoug
732.
by buildbot 3y ago
Exactly. Give us access the model and let independant researchers test it. OpenAI did this with GPT4, opening access publicly and deeper access to researchers within and outside of Microsoft. I simply don't believe the model is that go
733.
by buildbot 3y ago
I imagine they are, as well as the e2e encryption added to icloud after they dropped the CSAM thing. My pet theory is basically a rouge VP started the on device scanning project with whatever external org they worked with, and basically bli
734.
by buildbot 3y ago
This made me think - is a concept like most even defined for infinity?
735.
by buildbot 3y ago
I see there is a while keyword - doesn’t this make it turing complete? You can now loop, call functions, set variables - you could build a full processor emulator in pure css!
736.
by buildbot 3y ago
Yeah, and taking humans who hate bikes out of the equation - I basically stopped biking because several drivers literally tried to run me down (on a residential street no less).
737.
by buildbot 3y ago
I would hope so. UW CSE had fun names for every computer on their network, though my knowledge is dated to 2016. Humuhumunukunukuapuaa was eventually renamed because anytime you had to type it really sucked.
738.
by buildbot 3y ago
A massive reason why I, a consumer of Apple products, continue to consume them, new or used, is that I can trust that a repair will be genuine, be calibrated, and be as close to new as possible. After many years of Android phones that broke
739.
by buildbot 3y ago
Totally agree And now actually, I don’t even need service with satellite SOS.
740.
by buildbot 3y ago
Very interesting, but on my initial skim through, I don’t understand how this technique is lossless? Reproduced from the methods section: 1. Select a sequence of tc tokens that the model has already generated or completed predicting as the
741.
by buildbot 3y ago
Everyone should read the book! (The Cuckoo’s Egg) - it’s really awesome.
742.
by buildbot 3y ago
Well sure if you model is small and light enough. But there's no training a 7B+ model on one (well, you could it would be so, so, so slow). Like, decades?
743.
by buildbot 3y ago
Well, sometimes they fall of the back of trucks I guess: https://www.ebay.com/itm/134540730431 Archive link: https://archive.ph/7dPFo
744.
by buildbot 3y ago
They already have inference specialized designs and architectures. For example, the entire Jetson line. Which is inference focused (you can train on them but like, why would you?). They have several DNLA accelerators on chip besides the GPU
745.
by buildbot 3y ago
This 100% - plastic clothing is sadly just a bad idea. Also, smells gross very quickly. Wool, silk, cotton and other natural fibers are great and not hideously more expensive in most cases.
746.
by buildbot 3y ago
What's the alternative though? I don't like anti-competitive behavior at all, however thinking about what the regulation would be, I am just not sure there is any viable way. You have a few cases where the EU will go after whoever
747.
by buildbot 3y ago
This is absolutely not true for a vast part of the PNW. For example, with Verizon, I've bailed myself out of several situations where the alternative would be - wait (nobody showed up in the time we were waiting for help when we called
748.
Pixel owners report problems after installing January 2024 Google Play update(bleepingcomputer.com)
163 points by buildbot 3y ago|160 comments
749.
by buildbot 3y ago
I have sideloaded IPAs now, on iOS 17.3. Altstore.
750.
by buildbot 3y ago
20K amps (!!!): https://vimeo.com/853557623 They use probably the fanciest piece of rubber (and metal) ever made to pass the power in frontside to the die.

More ›