pdxww

6 ms·

1.
by pdxww 7y ago
Does it expose any sort of "device id"? Such ids are usually asked for by advertisers and iOS and Android gladly give it to them. I'm sure there are device-id to phone-number maps out there, and anyone with money can get acce
2.
by pdxww 7y ago
Maybe I don't understand something about these models. If the model was trained to mimic Trump tweets, it means that someone spent days of GPU time to find the weights of the model. Now if we want it to mimic HN comments, we'd ne
3.
by pdxww 7y ago
Regardless, this is a bad PR and they are losing potential customers. No one wants their car to be trimmed down just because the manufactures chose so. No, it doesn't matter what's written in the fine print.
4.
by pdxww 7y ago
The model needs to be retrained from sctratch for different types of texts. One can release a model trained to generate Trump tweets, but it's of not much use for generating fake news on a specific topic.
5.
by pdxww 7y ago
Luckily, uBlock Origin can block all portal tags entirely.
6.
by pdxww 7y ago
This sounds reasonable, but I think in practice the capacity of T1 won't be enough to capture long patterns and the F2 sequence is supposed to help T2 to restore the lost info about the longer pattern. The idea is to make T1 really goo
7.
by pdxww 7y ago
Can we diff spectrograms to define the "distance" between two chunks of sound and use this measure to guide the ML learning process? Would it help to decompose sound into subpatterns with Fourier transform? Afaik, there is a simil
8.
by pdxww 7y ago
T1 indeed contains all the info needed, but T1 also has limited capacity and can't capture long patterns. T1 would need to have 100s of billions weights to capture minute long patterns. I think this idea is similar to the often used sk
9.
by pdxww 7y ago
That's fine. The goal is to map "similar" 1 second chunks to similar vectors. I'm sure this can be done and uniqueness of sound won't be a problem.
10.
by pdxww 7y ago
I'm not an ML guy, so can't say if this is an autoencoder. We can combine multiple sequences in any way we want. Obviously, we can come up with some nice looking "tower of lstms" where each level of that tower processes
11.
by pdxww 7y ago
1 second of sound. Or a few seconds of sound.
12.
by pdxww 7y ago
We can think of a ML model that takes 1 second of sound as input and produces a vector of fixed length that describes this sound: S[0..n] = the raw input, 48000 bytes per second of sound F[1][k..k+48000] -> [0..255], maps 1 second of sou
13.
by pdxww 7y ago
The same way we map words to vectors or entire pictures to vectors. We'll have another ML model that would take 1 second of sound as input (48000 1 byte numbers) and produce a say vector of 128 float32 numbers that would "describe
14.
by pdxww 7y ago
To illustrate more this idea, let's use soundtrack v=negh-3hi1vE on youtube. Such soundtracks consist of multiple more or less repeating patterns. The period of each pattern is different: some background pattern that sets the mood of t
15.
by pdxww 7y ago
Unheard of. This is why in the US we have the 1st amendment backed by the 2nd amendment. The freedom of speech can't defend itself. What needs regulation is ISPs and companies that deal with personal data.
16.
by pdxww 7y ago
That CodeNet would be the SkyNet, essentially. What's shown here looks impressive, but it's the same good old text generator that can produce something that looks very similar to the dataset used to train it. It can't go be
17.
by pdxww 7y ago
Impressive. Would this model benefit from something like "dilated attention"? Instead of feeding it raw sound samples, we could split the input into 16 sec, 8 sec, 4 sec and so on slices, assign each slice a "sound vector&quo
18.
by pdxww 7y ago
If this idea worked in practice, go-ipfs/js-ipfs wouldn't have so severe problems with connectivity.
19.
by pdxww 7y ago
How could you possible make such a DHT? We live in the world of NATs, especially symmetric NATs, where each mobile phone user gets assigned a random ip:port every time it makes a connection. DHT, on the other hand, needs every node to have
20.
by pdxww 7y ago
It would. As it was mentioned earlier on HN, a woman can get 100k in tax-free alimony payments from a guy making 250k. She can even get his house.
21.
by pdxww 7y ago
This is not a roadmap, but rather a wishlist. There is a fundamental problem that IPFS needs to solve first. This problem is called an efficient WebRTC-based DHT. In order to change the web, IPFS needs to become usable in the browsers. Sinc
22.
by pdxww 7y ago
An airplane usually spends 1-2 hours in the airport between flights. During that period it needs to be recharged.
23.
by pdxww 8y ago
If we look at this from a buddhist's point of view, then there is no luck: the author did something in past that allowed him to use this opportunity and what he did could be completely unrelated to his work, e.g. he wrote a blogpost lo
24.
by pdxww 8y ago
Honestly speaking, since it's an anonymous forum, what's the difference between L5 and L7? Like if I was asked the same question - the difference between L3 and L5 - I'd say that it's just experience with a few interesti
25.
by pdxww 8y ago
And this is irrelevant also, because we can introduce sampling the way I described and can continue driving just fine.
26.
by pdxww 8y ago
This is smart. Unethical, but smart.
27.
by pdxww 8y ago
It's great that eyes can detect polarization and orbital momentum. My point is that none of that info is needed to drive a car.
28.
by pdxww 8y ago
"Senior manager" isn't a remarkable achievement. It's just a role or position in a company that needs to be filled in. She happened to be the one who got picked (for various reasons). If she founded a company with $1B pr
29.
by pdxww 8y ago
A simple abstraction here: we put two high-quality monitors before our eyes and get all the data from them. Can we continue driving this way? Absolutely. This means that no matter how our eyes actually work, an array of RGB pixels is enough