vonnik

84 ms·

901.
How to Speed Up Spark with Native Binaries and OpenBlas(deeplearning4j.org)
1 points by vonnik 11y ago|0 comments
902.
by vonnik 11y ago
I think we already migrated. Right now, each story is a bundle of facts and statements. A chatbot could disaggregrate those facts, and feed the news consumer answers to questions they have about the subject without imposing such a total ord
903.
by vonnik 11y ago
Former NYT journalist here. I'm really glad the paper is experimenting like this. I want to make a slightly bolder statement than the headline of the post: The future of news is not written, and it's not one way. It's convers
904.
by vonnik 11y ago
I love posts like this, but I suspect another set of algorithms is going to win this race. Probably some combination of convnets and recurrent neural nets. http://www.unnvision.com/pub/DDNN_ECCVW2014.pdf http:/&#
905.
by vonnik 11y ago
stream-of-consciousmess ;)
906.
by vonnik 11y ago
I think it's less about the head getting chopped than about having "the head take up about 1/3 of the image," as Karpathy says. So what the net is learning is composition, or balance in an image, which is really cool. Th
907.
by vonnik 11y ago
The PG headline is really the first half of a larger sentence: Write like you talk, and try to talk in concise, plain language. If you have trouble doing this, the Hemingway App might help. http://www.hemingwayapp.com/ Ther
908.
by vonnik 11y ago
i never worked for a startup i didn't like, which is lucky i suppose. i left futureadvisor earlier this year to found my own company, but if i hadn't felt that itch, i would have stayed working with bo lu and jon xu for many years
909.
by vonnik 11y ago
I spent about 12 years as an employee in companies of various sizes in different industries. Purely anecdotally, I can tell you that sh*t rolls downhill. Bad people at the top make for unhappy teams and poor performance. The opposite is als
910.
by vonnik 11y ago
While this automated solution appeals to technical people, who would like to get past the messy relational approach to PR, it's a false promise. PR doesn't really scale. Like recruiting, it's just hard work. You're deali
911.
by vonnik 11y ago
It's more complicated than that. You're extrapolating that because they put their email up on the site, they sent the signal that they want these spam-blast form pitches. That's a logical leap, and it happens to be incorrect.
912.
by vonnik 11y ago
Right. And I'm not saying you need to know them personally because they partied on your yacht. I'm saying you should do your research. Study them. Know who they write for and what they like. Anyone can do that. Based on that knowl
913.
by vonnik 11y ago
And I would argue that the effect of porting out-of-date articles is approximately the effect startup founders will achieve when they send form pitches to reporters. The reporters will know that something is off.
914.
by vonnik 11y ago
This is simply not true. Some do. Most already get too many pitches. And the crucial point that we're not addressing here is: Pitched about what? A lot of work has to be done to target and tailor pitches, and that can't really be
915.
by vonnik 11y ago
I made these same points below. https://news.ycombinator.com/item?id=10440046 It's all about knowing the reporters. Form emails can't address that.
916.
by vonnik 11y ago
There are a lot of misconceptions about the press in this article, and some outdated information. First, let me just say I spent a decade working as a reporter for the NYT, the IHT, Bloomberg, Businessweek and others. For the last two years
917.
by vonnik 11y ago
that makes sense. thanks. EDIT: followup question: to what extent does DeepMind's Deep-Q learning address the problem of associating remote causes with final outcomes?
918.
by vonnik 11y ago
Maybe I'm thinking about things wrong, but can't you have a scalar quantity to represent error for backprop at the end of a neural network (in a supervised regression problem for example)? That scalar error becomes a vector as it
919.
by vonnik 11y ago
Yes, I should have added the context about the scalar reward, but I'm not sure why that changes anything. This may seem like a naive question, but it's sincere: What makes a scalar reward less effective at modifying a Q function t
920.
by vonnik 11y ago
"There is no way in hell that you can learn billions of parameters with RL." I really love LeCun's provocative stances, but I get suspicious when people talk about impossibilities. RL is making huge strides. People adopted th
921.
by vonnik 11y ago
One of my economics professors in undergrad used to say, when teaching Milton Friedman's monetary theory: "Given these assumptions, only GOD could reach a different conclusion." And then he would cackle absurdly. The point wa
922.
by vonnik 11y ago
We can engineer yeast to produce almost any molecule. We can synthesize meat. We can do that with nearly free electricity from solar, among other ingredients. These things are happening now. Arable land is an archaic fetish, like much of St
923.
by vonnik 11y ago
These guys are jokers. A security firm that gets pwned by Anonymous... http://venturebeat.com/2011/12/28/anonymous-stratfor-hack-10... Stratfor is flattering the sensibilities of the military industrial com
924.
Neural Networks with Few Multiplications (Bengio and Co.)(arxiv.org)
3 points by vonnik 11y ago|0 comments
925.
by vonnik 11y ago
AKA robots that replaced human labor.
926.
VR in a Sensory Deprivation Tank(youtube.com)
1 points by vonnik 11y ago|0 comments
927.
Deep Learning Use Cases(slideshare.net)
2 points by vonnik 11y ago|0 comments
928.
Resources to Start Using Deep Learning(deeplearning4j.org)
2 points by vonnik 11y ago|0 comments
929.
Can robots be creative?(theguardian.com)
1 points by vonnik 11y ago|0 comments
930.
by vonnik 11y ago
I used to work for a financial news service. The Fed's interest rate announcements were automated; they hit the wire service, which had an automated template to push out headlines on the news (human oversight but not much intervention)

More ›