mirker

14 ms·

151.
by mirker 5y ago
If you have customers, does that mean the incremental gain from an improved index costs too much to store? Or are you talking about computational costs?
152.
by mirker 5y ago
Yeah, PhD income is $40k or less a year. So 5x above market rate for high-risk research. On top of that, PhD students are well-aligned for the research and have other non-monetary reasons to pursue the research (their PhD requirements).
153.
by mirker 5y ago
My advice for high school and lower level college maths is to 1) pick a good book 2) solve all the problems. A good way to accomplish that well is to start at the lowest math that is possibly non-trivial for you. Otherwise, you will hit a w
154.
by mirker 5y ago
By “backend” I mean the compiler logic backing traced tensors in JAX. torch.fx seems different in that it’s primarily aimed at being a platform for users, rather than JAX which (as far as I know) hides the logic behind @jit annotations. Bot
155.
by mirker 5y ago
My read on it is it’s equivalent to the backend of JAX’s @jit. You can write your own @jit to do operator fusion and other optimizations.
156.
by mirker 5y ago
[And a program/computer may not think yet still does interesting computations]
157.
by mirker 5y ago
I just looked for Drupal and Wordpress jobs and they exist ($60k/year seems normal). Since the whole point of these frameworks is to abstract out actually writing the bulk of server code, I’d imagine a large part of the job is dealing
158.
by mirker 5y ago
Being a prodigy is not necessary to set up a webpage (or whatever). I’d argue most jobs are processing low-grade data with legacy frameworks, which is a test of patience and not prodigy IQ.
159.
by mirker 5y ago
The carbon cost of producing a car is the same as the cost of driving it [1]. Sure, for the sake of argument, you can say that each person has 1 new car per lifetime, but it’s really hard to say definitively how all the carbon is being mov
160.
by mirker 5y ago
To me it seems like there was way more leverage from Apple’s side, so the “negotiation” was really a nicety, which was subsequently reverted. The company was eventually sold at a loss rather than breaking even, so the true market value was
161.
by mirker 5y ago
You’re ignoring the economies of scale of the human civilization. A car comes from a factory which services millions of humans, so there isn’t any reason to believe that it’s as simple as scaling individual usage. There are a ton of polluti
162.
by mirker 5y ago
There are also a lot of indirect carbon costs. Doing the calculation accurately is a recursive time-series analysis where you have to trace how every object was used to create other objects through the whole earth’s supply chain. So having
163.
by mirker 5y ago
I don’t know much about acquisitions, so what is the incentive of the acquisition from the buyer’s side? Apple and Facebook seemed to create clones of the functionality pretty easily, so the main purchase benefit seems risk over implementat
164.
by mirker 5y ago
Here’s another way to put it in terms of money: * Research universities make money from research and not teaching. * Admins create promotion schedules on research and not teaching. * Teaching is the “cherry on top”. Good research > good
165.
by mirker 5y ago
I personally compare cooking time to time to go out end-to-end . Even with fast food, you have to drive to the pickup window, wait for the food, etc. It’s easily 10 minutes and you have to do it 1-2 times a day, so 2-3 hours a week. I’ve h
166.
by mirker 5y ago
Like prescription medications, which fuel the opioid crisis?
167.
by mirker 5y ago
I am quite confused. I thought Windows 10 “fast boot” was hibernation and default “shutdown” behavior. I thought Windows 10 startup times were faster because of hibernation shortcutting the real boot sequence.
168.
by mirker 5y ago
I mean it’s a bit murkier than that: 1. GPUs used to be much less programmable and had fixed-function graphics pipelines. 2. General purpose GPU computing was demonstrated a decade before the 2012 deep learning craze. OpenCL/CUDA was q
169.
by mirker 5y ago
Yeah, the motivation is pretty easy if you consider that much of software engineering is the same process (e.g., performance tuning), and on top of that you get prestige and do some math on the side. It’s basically a research version of tes
170.
by mirker 5y ago
Thank you for the comment and I wish you luck. Yeah, I remember organic carrots were hip among past engineers a few years ago. I’d imagine it’s pretty hard to figure out profit projections without lots of years of data. I think apprenticesh
171.
by mirker 5y ago
How does the effort pay off compared to a “normal job”? I get the fulfillment aspect, but is it cost effective for the labor? To give an example, I can buy a few pounds of potatoes for $5, and I cannot possibly see how I could do better tha
172.
by mirker 5y ago
How do you know if the building is structurally safe? I’m all for DIY in general, but with respect to a house, my biggest concern would be having the second story of the house collapsing into the cellar :) And yes, my grandparents in Europe
173.
by mirker 5y ago
The diagram has 3 input companies and 3 output companies, so at a surface level it seems nothing has changed. Even Google Fiber couldn’t compete with the big guys because (at least partially) these utility companies are so entrenched in lob
174.
by mirker 5y ago
AT&T just re-merged after the break-up, so it’s a bad example. See the diagram: https://www.theverge.com/2016/10/24/13389592/att-time-warner...
175.
by mirker 5y ago
How hot does that get at full load? I was worried about having so many cores at factory if not overvolted. For example, you can find plenty of people freaking out about 5800X temps. On the other hand, people claim the high temps are fine.
176.
by mirker 5y ago
I believe the applications are mostly what you describe. You take a very general algorithm like a NN or raytracing and scale the model and/or data. For NN, it’s quite easy to use a huge amount of memory by making back-propagation chain
177.
by mirker 5y ago
Nit but bandwidth isn’t related to memory capacity. Bandwidth is a function of cache/access pattern and is independent of total memory size.
178.
by mirker 5y ago
Yes, but it’s meant to do ML inference, which can be parallelized decently. On those workloads, you can use GPUs, which are also composed of thousands of “wimpy” cores.
179.
by mirker 5y ago
Related: a common mistake is treating foods as binary “healthy” or not. Nuts, for example, can be “healthy” but ten handfuls is probably excessive.
180.
by mirker 5y ago
The food-pyramid is honestly creepy. First, the order of foods is determined by invested corporations. Second, it’s arranged in a convenient shape (triangle) like some poser illuminati cult. Third, it’s taught to kids when they don’t know a

More ›