glup

23 ms·

181.
by glup 8y ago
Because I want it to be around in two years.
182.
by glup 8y ago
Try https://www.nytimes.com/section/todayspaper , should be ad free for subscribers (it's a closer approximation to the actual print version)
183.
by glup 8y ago
Cool platform/capabiluties aside, leave it to Uber to figure out how to make an owl into a really creepy logo.
184.
by glup 8y ago
Commercial APIs from leading companies generally achieve better performance, but besides obvious price and network latency, they are complete black boxes so you can't diagnose and fix problems.
185.
by glup 8y ago
The factors in the data aren't just your voice but recording environment and microphone, and a recognition "engine" is really a combination of model architecture (sometimes separate acoustic and language models), training dat
186.
by glup 8y ago
I think this is a great summary of where we are at now, but I think stronger broad-coverage language models (=expectations for what people say, better generative models of speakers) are feasible (for a couple 10s-100s of millions in R&D
187.
by glup 8y ago
If that quote attributed to Fred Jelineck ("Every time I fire a linguist, the performance of the speech recognizer goes up") is right, they should have negative linguists.
188.
by glup 8y ago
In my experience DeepSpeech with the default Mozilla model + 5-gram is extremely sensitive to background noise. Fiddling with the hyperparameters, e.g. upweighting the language model, helps a little.
189.
by glup 8y ago
What probabilistic generative model of language generates ``Balls have zero to me to me to me to me to me to me to me to''? Is this a bigram model taking the highest probability continuation? Certainly nothing modern, right?
190.
by glup 8y ago
Have you tried reading the 2-3 "meat" articles in the middle of the New Yorker each week? I find that these 8-15 page articles are hugely informative in comparison to the Associated Press funnel format and clickbait drivel that do
191.
by glup 8y ago
I feel like I would have been a lot more interested in calculus in high school if I had first understood how statistics could be used for inference, and how calculus was vital for understanding statistics.
192.
by glup 8y ago
out of interest, what city?
193.
by glup 8y ago
I am surprised human biomass is so much among animals, just 1:16 against arthropods? And 30:1 against birds?
194.
by glup 8y ago
I've come to the conclusion that better documented, quality controlled parts are worth it as my free time for hardware projects dwindles. In the end the parts for RPi and Arduino projects are generally quite cheap in comparison to my t
195.
by glup 8y ago
I didn't get it until I read your post out loud.
196.
by glup 8y ago
1) Buy a Raspberry Pi and an Arduino 2) Pick a project: telepresence robot, autonomous robot, sous vide machine, thermal camera trigger, etc. 3) Buy minimal pieces for that project: AdaFruit, SparkFun, or various OEM pieces from Amazon
197.
by glup 8y ago
The difference is that the people in academia know that it is statistically unsound and choose to act unethically. I think the problem in A/B testing is that a lot of developers don't know it is unsound.
198.
by glup 8y ago
This is a common practice in a number of other fields (e.g. cognitive science and psycholinguistics), at least for conference submissions. In general I don't see a huge difference -- when a grad student has sufficient standing and expe
199.
by glup 8y ago
I think there's a confound there that these are both very thin machines, limiting the size of fans (=CFM per an acceptable noise level) and heatsinks. Not sure how others feel but I wouldn't mind a slightly larger form factor if i
200.
by glup 8y ago
I understand this variability in thermal efficiency in who-knows-who-made-this-but-it-sure-was-cheap OEM ecosystems, but shouldn't Apple of all companies have the engineering/sourcing/testing resources to get this down to +-1
201.
by glup 8y ago
OP probably meant POSIX-compliant
202.
by glup 8y ago
That should work for revealing throttling -- you will see that it tops out at 1.1 GHz or something.
203.
by glup 8y ago
It’s interesting to note that there are very few repeated answers or “me too” type posts, which suggests that what works best is highly idiosyncratic and depends on a lot of factors (what you need to organize, with whom, and in what ecosyst
204.
by glup 8y ago
https://github.com/Daij-Djan/nocturne
205.
by glup 8y ago
I actually use Nocturne pretty frequently to turn my screen monochrome if I am doing stuff outside of iTerm or my text editor — I find the use of color unmotivated and distracting in most programs, and especially websites
206.
by glup 8y ago
This. It has always seemed pretty nuts to me the slippage between the sort of idealized multi-armed bandit mechanics of A/B testing (the theoretically sound basis) and the actual real-world situations with enormous hypothesis spaces an
207.
by glup 8y ago
This is very hard to estimate -- 1) the oil companies may find new sources 2) they may find that existing beds aren't as productive as projected 3) extraction methods could improve 4) the rate at which we are using it could increase or
208.
by glup 8y ago
That's a cool appendage, but there are still a lot of problems in grasping / planning with materials like fabrics, right? e.g. this doesn't solve the notoriously hard laundry folding problem
209.
by glup 8y ago
Why is an article like this okay, but you'd never see a clickbait article for illegal activities called "pros and cons of embezzlement"? Is stealing fine art not really stealing?
210.
by glup 8y ago
You can run those on the remote side of the mosh session

More ›