beecafe

9 ms·

31.
by beecafe 4y ago
This paper came out after. You read the date wrong. The weird formatting and very short size also point to this. This paper is a clear rip-off and an instance of academic fraud.
32.
by beecafe 4y ago
This paper is a bad copy of MLP-Mixer.
33.
by beecafe 4y ago
Or it could be that people are LGBT and can now safely come out and be LGBT? If you look at a chart of handness, as left-hander were no longer forced to be right-handers, the proportion of left-handers shot up till stabilizing at the natura
34.
by beecafe 4y ago
It seems like you could 3D print all of it
35.
by beecafe 4y ago
You can just find the most matching snippets in the training data using an embedding model.
36.
by beecafe 4y ago
those APIs would be worthless if nobody wanted to stitch them together
37.
by beecafe 4y ago
IPFS is this
38.
by beecafe 4y ago
Radio telescopes don't care about the atmosphere. And only Arecibo was powerful enough to send messages for SETI, or do radar imaging of solar systen bodies.
39.
by beecafe 4y ago
Its the desktop and laptop users who will go extinct. Changes which bother them really don't matter at all.
40.
by beecafe 4y ago
https://en.m.wikipedia.org/wiki/Kademlia
41.
by beecafe 4y ago
nitpick: it's not inverse fourth power, it's just inverse square with twice the distance, which is 1/((2 * distance) ^ 2) = 1/(4 * distance ^ 2) (ignoring reflection efficiency, but that isn't determined by distance
42.
by beecafe 4y ago
You don't even need to do a diff, the model itself actually predicts the diff (which is how it samples the image) so you could just stop once the model is predicting close to 0
43.
by beecafe 4y ago
Each sampling step runs at a specific scale, fewer steps would skip some of the intermediate scales
44.
by beecafe 4y ago
You could compile the rust version to WASM and run it in browser.
45.
by beecafe 4y ago
You are years behind if you think you're training a model worth anything on consumer grade GPUs. Table stakes these days is 8x A100 pods, and lots of them. Luckily you can just get DGX pods so you don't have to build racks but for
46.
by beecafe 4y ago
Everything takes time. Something has (exchange) value if you get given something of value in exchange for it. Something has (intrinsic) value if you value that thing itself to give value in exchange for it, even if nobody else would give it
47.
by beecafe 4y ago
You're talking about second order optimization. The full-fat version is too slow (the Hessian, the matrix of second derivatives, is huge for DL models), the optimized approximations don't seem to provide a meaningful benefit over
48.
by beecafe 4y ago
Software is trivial to make, it should rightly have no value.
49.
by beecafe 4y ago
Or it could just be that all UFO sightings have been established to be human made and militarily relevant, and this provision allows them to handle that without Congress getting in the way.
50.
by beecafe 4y ago
It would be great if consumer electronics products had expected lifetime ratings on the front. People would naturally take issue with products with significantly below median lifetime, and also if their own products didn't live up to t
51.
by beecafe 4y ago
As they should be, natives clearly have lower average merit otherwise they would immigrate somewhere where they could easily get funding.
52.
by beecafe 4y ago
You're right, I misremembered, it was for equality instead.
53.
by beecafe 4y ago
ML still needs runtime type checks due to references
54.
by beecafe 4y ago
Geoffrey A. Landis, "Prospects for Solar Pumped Semiconductor Lasers," Paper SPIE 2121-09, Laser Power Beaming, SPIE Proceedings Volume 2121, pp. 58-65, January 27–28, 1994
55.
by beecafe 4y ago
You could build a solar pumped laser that uses sunlight shining onto iodine to directly make very intense laser light.
56.
by beecafe 4y ago
How does it work? I can imagine a program that checks if it sees a correctly signed message of the right time, but that program would then have to decrypt the payload, and I don't know how the key for the payload could be derived from
57.
by beecafe 4y ago
Sounds similar to ML research
58.
by beecafe 4y ago
https://youtu.be/qg8pMUd-tSk You can make the OLED part at home, but patterning it into a screen with an active matrix of see-thru transistors is hard.
59.
by beecafe 4y ago
You should take the first step and kill yourself. Then we can talk about what others should do.
60.
by beecafe 4y ago
So you'll go over and review your own browsing with them too right?

More ›