blt

74 ms·

871.
by blt 8y ago
I totally agree, it's kind of depressing. With deep learning we had an opportunity to pick a new language because ppl would have used whatever language was necessary to get what tf (& theano) provide. If Google had written tf in Ju
872.
by blt 8y ago
Do you have details? I noticed that Youtube on Firefox makes the fan spin up on my 2013 rMBP a lot. I currently use Firefox but as a heavy Youtube watcher this is a pretty big negative.
873.
by blt 8y ago
Same. Ghostery on Android
874.
by blt 8y ago
YAML and TOML both seem too complicated. Automatic date parsing? So many different ways to specify the same nested hash table? I like json because there's usually one obvious way to do what you want. It's a local minimum, like C a
875.
by blt 8y ago
Yes
876.
by blt 8y ago
This is a good idea, could be really popular with semi-technical users.
877.
by blt 8y ago
I agree. I've given talks on pretty math-heavy ideas but they can still usually be explained visually in a way that is significantly easier to understand without losing fidelity. I love LaTeX and Beamer for pure text/math, but onc
878.
by blt 8y ago
This is the first time I've heard anyone complain about GLSL syntax. (I am only a dilettante in gfx.) What is the problem? It seems pretty well suited to the task in my opinion.
879.
by blt 8y ago
The highlighted sections on this page drive me crazy. They really interrupt the flow of the text.
880.
by blt 8y ago
There has been a lot of research into using very low bit depth weights in neural nets, pruning, etc. I am pretty confident that this research, combined with purpose-designed silicon, will allow us to evaluate quite powerful neural net model
881.
by blt 8y ago
Frameworks != reproducible results. Ideally, they'd release all the code you need to get the results from the DDPG paper. I recognize that the code probably depends on proprietary DeepMind tools, making this not feasible. Still, it suc
882.
by blt 8y ago
TBH, I would say the ratio of code published to papers published for DeepMind is pretty small. They have generated a lot of non reproducible results.
883.
by blt 8y ago
In Deep Q Network (DQN), the Q network itself is the policy. You have one output for each possible action, and the the neural network estimates the Q value for each action in the current state. You act by selecting the action with the highe
884.
by blt 8y ago
Yeah I've noticed that pattern when I implement numerical algorithms by hand. A lot of the time they're pretty simple. But you need the analysis to be confident that it will work. A point in favor of formal CS / applied math
885.
by blt 8y ago
The jewel of monte carlo methods is Markov Chain Monte Carlo (MCMC) for sampling from high-dimensional probability distributions. The math is more involved, but the problem it solves is really hard and really important.
886.
by blt 8y ago
Because you have observed that emojis do not help you communicate, and then concluded that emojis cannot possibly help anyone communicate. There are lots of people in this thread who have mentioned concrete examples of emojis "prov
887.
by blt 8y ago
I suffer from social anxiety. There were a lot of times in the past when I would receive a text and fail to reply because I got stuck trying to figure out how to express myself "correctly". Emojis and the normalization of emoji-he
888.
by blt 8y ago
Why are you so intent on invalidating other people when they say that emoji help them communicate?
889.
by blt 8y ago
Have you ever encountered difficulty conveying or understanding conversational tone over the internet? No? You are lying or lack self-awareness. Voice and body language are important for disambiguating sentences with more than one possible
890.
by blt 8y ago
If there isn't one already, I'm sure someone will implement OpenGL on top of Metal when it's needed badly enough. At least they're going closer to the hardware, not further away.
891.
by blt 8y ago
> You will not be able to search the code in a fork that has less stars than its parent. ... but since it's allowed if the fork has more stars, I guess they're just trying to save compute cost. Seems like a bad place to scrim
892.
by blt 8y ago
It is shockingly bad. The fact that you can't code search in a fork blows my mind. How have they not fixed this basic, important feature after so many years? What could possibly make it more difficult than a few person-months of effort
893.
by blt 8y ago
The algorithm producing the EchoNest scores has some questionable results, like Hollaback Girl - 38% acousticness and Hey There Delilah - 65% danceability.
894.
by blt 8y ago
I agree, the GDC talk is extremely inspiring.
895.
by blt 8y ago
yes. ownership is an extremely critical part of software design, especially for complex interactive programs.
896.
by blt 8y ago
I guess you are getting down voted for lack of content, but I agree. Those are all very bad "Java without GC" C++ practices that lead to awful code bases. Thankfully the author seems to have learned that from their experience writ
897.
by blt 8y ago
Reading this thread, it seems like there are doubts about whether this particular product will really work like non-volatile RAM. Either way, it's an indicator that the real thing will be here soon. I can't wait to see how fast pe
898.
by blt 8y ago
but they also seek publications in the main Nature, which is unusual in the field.
899.
by blt 8y ago
> You often see a whole chapter in the beginning devoted to explaining the philosophy behind the technology, rather than letting the user get a feel for the philosophy themselves, by doing. I agree that this is annoying. As an author,
900.
by blt 8y ago
IMO, the landing page of any programming language site should include some code samples demonstrating what makes the language different from the crowd.

More ›