rryan

8 ms·

61.
by rryan 10y ago
You're incorrect.
62.
by rryan 10y ago
I think I will never understand the sort of mindset where you just say things regardless of whether you know them to be true. https://www.google.com/maps/place/Google+UK/@51.5246616,-0.1...
63.
by rryan 10y ago
No, it's correct (new in Python 3.6): https://www.python.org/dev/peps/pep-0498/ The print statement however is not valid Python 3 :).
64.
by rryan 10y ago
How many people using RethinkDB are changing its code? AGPL does not mean you have to share the code to your app making use of RethinkDB. AGPL protects against people making hosted forks of RethinkDB and not sharing the changes they've
65.
by rryan 10y ago
To be fair -- Twitter is a much harder architectural problem (significant fan-out and fan-in problems to solve). Pokemon Go is less complicated in terms of the conceptual model. There are only a few pretty simple ways that different players
66.
by rryan 10y ago
I've seen this sentiment in a couple of places now -- and the news media and non-technical folks seem to refer to "encryption" very loosely to mean "protected in the way I want it to be protected at the times I want it t
67.
by rryan 10y ago
This seems to be where the "backing off" claim is coming from: http://www.theverge.com/2016/5/18/11699122/google-allo-messa... > First, all conversations are encrypted "on the wire,&quo
68.
by rryan 10y ago
Unfortunately no, see Amdahl's Law. https://en.wikipedia.org/wiki/Amdahl%27s_law
69.
by rryan 10y ago
So cool -- can't believe he also made the lens!
70.
by rryan 10y ago
> Probably because that's not what it was designed for. TF was explicitly designed with distributed training in mind (their initial whitepaper and the DistBelief paper that came before it make this clear) -- I don't know how yo
71.
by rryan 10y ago
My dotfiles repo includes an "initialize_debian_server.sh" that basically does all this. It's definitely doable.
72.
by rryan 10y ago
> It could be a great area of research on how to reduce the amount of data it takes to train the network. This is typically referred to as "one-shot learning" in the literature -- and people certainly work on it!
73.
by rryan 10y ago
This site is so handy -- if it's useful to you too please Flattr the author so they can continue to host it. It's been a helpful tool to be able to link people to code+assembly when trying to convince people not to "optimize&
74.
by rryan 10y ago
There's a reason Google has an absolutely massive office in Zurich. Anecdotally, some of the sharpest Googlers I have ever collaborated with are in Zurich.
75.
by rryan 10y ago
> Generating random patterns that sound jazz-ish is interesting, but until multiple generators can react to what the other is doing in real time (or to a human participant), it isn't exactly jazz. For an arbitrarily complex network,
76.
by rryan 10y ago
I think then the question becomes, are humans doing anything different than that? If so, why do you believe the network is only reproducing statistics rather than having learned the same circuitry humans have when improvising/composing
77.
by rryan 10y ago
This is really neat! But I think it's a stretch to call it AI-generated jazz music. As I understand it, the author has trained an LSTM on a single MIDI file -- "And Then I Knew" by Pat Metheny. The network is then asked to ge
78.
by rryan 10y ago
Let's say it was -- how could you, the end user, go about evaluating that claim? Even if the perfect technical solution existed, it wouldn't fix this problem.
79.
by rryan 10y ago
No, that doesn't work. c.f. Lavabit and "Reflections on Trusting Trust"
80.
by rryan 10y ago
It's on the page titled "Roadmap", so ... yes? https://github.com/tensorflow/tensorflow/blob/master/tensorf...
81.
by rryan 11y ago
While you're right that Google's implementation of Content ID doesn't handle fair use well, Content ID is critical for indie creators. Facebook's lack of protections lead to tons of lost revenue for them (and for indies,
82.
by rryan 11y ago
What do people think of this video? https://www.youtube.com/watch?v=t7tA3NNKF0Q IMO Content ID is essential for a healthy video creator ecosystem. Google's implementation may give the big media companies too much power
83.
by rryan 11y ago
This video claims that Facebook's lack of a content ID system is harmful to individual / small content creators as their videos are copied and go viral with no compensation to the content creator. https://www.youtube.co
84.
by rryan 11y ago
Also, this is t-SNE: https://en.wikipedia.org/wiki/T-distributed_stochastic_neigh... The S is for "stochastic" -- i.e. you get a different 2D projection every time you run it on the same inputs. Take it with
85.
by rryan 11y ago
It's quite different from the original Flume paper as Dataflow supports expressing streaming computations in addition to batch in a unified set of APIs. It's like a combination of Flume and Millwheel.
86.
by rryan 11y ago
Glad you like it :). We're working on the last mile stuff right now (updating the manual, website, etc.) -- it will definitely be out before new years!
87.
by rryan 11y ago
Mixxx developer here -- can confirm our VBR decoding code is slightly less reliable than CBR! It's easier when you know exactly how many samples you're going to get out of a frame. We rewrote a bunch of this in 2.1 (coming in Q1 2
88.
by rryan 11y ago
Here's one incredibly awesome feature of Clang: Windows support (thanks to Google, actually). No Cygwin/msys required -- it implements the Windows calling conventions, exception handling, etc. You can link modules compiled with Cl
89.
by rryan 11y ago
> and this benchmark seems to back his claims. These benchmarks evaluate single-node performance. LeCun's remarks were concerning distributed training (specifically that bandwidth between machines is a limiting factor to scalability
90.
by rryan 11y ago
Great work -- very exciting to see a Rust ML framework! Dist-Belief (TensorFlow's predecessor) was oriented around layers (as Leaf is) but TensorFlow (like Theano) is graph/flow-based rather than layer-based. I think in particular

More ›