Veedrac

67 ms·

751.
by Veedrac 7y ago
> Nobody has even made serious efforts to use deep learning for physical design DeepMind have for place&route IIRC.
752.
by Veedrac 7y ago
It's model parallel, so the first thing you do is lay out your floorplan for the model, which looks like this. https://secureservercdn.net/198.12.145.239/a7b.fcb.myftpuplo... Then you put your data next to the cor
753.
by Veedrac 7y ago
The appropriately large models with public recognition I know of use attention, which is too memory-hungry to work effectively on the CS-1. The datasets aren't the issue. I'm fine with skepticism. It's certainly plausible tha
754.
by Veedrac 7y ago
A bit baffled by this because on every axis I look this seems like a dream of a compilation target. * No DRAM or caches, everything is in SRAM, and all local SRAM loads are 1 cycle. * Model parallel alone is full performance, no need for da
755.
by Veedrac 7y ago
They have customers already, one (Argonne National Labs) is given explicitly. The issue with using ‘industry standard’ benchmarks is that it's like measuring a bus' efficiency by shuttling around a single person at a time. The CS-
756.
by Veedrac 7y ago
If anything this post disagrees with the claim that ‘English isn't generic for language’. It considers a naïve approach that fails for English, and shows that it fails for Russian in a similar way, and then compares it with an improvem
757.
by Veedrac 7y ago
My claim isn't that it's easy to solve NLP; of course a resource-constrained system like Google Translate that inderects many language pairs through English is going to have glaring issues. My claim is that the challenges generali
758.
by Veedrac 7y ago
An NLP model that can't even (implicitly) segment CJK languages is laughable. It's like saying ‘sure, Magnus Carlsen is really good at chess, but in draughts you can take multiple moves in a row.’ If you can handle the ambiguities
759.
by Veedrac 7y ago
They're all similar to English. Languages have their syntactic ambiguities in different places, but they're all fundamentally similar . In every case the hard part is in understanding the semantics expressed, and the characters u
760.
by Veedrac 7y ago
This is simply because Google Translate relies on language-language models trained on language-language pairs. Where training data is not available in quantity, the quality of translations is sufficiently low that going through English is p
761.
by Veedrac 7y ago
I'll agree that if future designs look like ARC, and cost $5-6B, fusion won't be economical.
762.
by Veedrac 7y ago
People have priced these already, it's competitive. Fusion requires fewer safeguards, less insurance, and is more repairable than fission. Reliability isn't an issue just because it's 4m³. Why would fusion reactors need ‘very
763.
by Veedrac 7y ago
OK... so? This seems like a pretty meaningless statistic. The reactor core for a magnetic confinement fusion reactor is pretty much the only meaningful radioactive waste from the process, and within 50 years it turns itself into low-level
764.
by Veedrac 7y ago
NuScale is about 3x3x20m for 60 MWe. SPARC is about 4x4x4m (?) for about 50-100 MWe; that only includes the reactor proper, but the rest should be no less compact than any other plant.
765.
by Veedrac 7y ago
First Light Fusion, Commonwealth Fusion Systems, and Tokamak Energy are all aiming at net energy by 2025 and grid-connected power by 2030. Lockheed Martin wants power plants at scale in the 2020s, with designs small enough to power aircraft
766.
by Veedrac 7y ago
His graph assumed continuous exponential growth in deployment. In reality, deployment of solar in 2018 (97.4 GWdc) was slightly less than in 2017 (97.8 GWdc). > But anyway, is that really a talk saying we should invest in fusion becaus
767.
by Veedrac 7y ago
The claim is that a renewable grid would be manageable and affordable IF we could produce enough solar and wind installations, but that even optimistic projections don't have enough supply (of solar/wind/etc. installations) t
768.
by Veedrac 7y ago
> I'd love to see some numbers here because without them it's easy to assume this is not cost competitive with solar+battery in +5 years given current trends. [On nuclear fusion] “It's kind of as instructive to ask what&#x
769.
by Veedrac 7y ago
That's the point of the hard version.
770.
by Veedrac 7y ago
Only whichfaceisreal.com uses the original StyleGAN. The github.io links use StyleGAN2.
771.
by Veedrac 7y ago
I thought StyleGAN2's teeth were one of the biggest upgrades to the face proper. The original's were pretty bad, but removing the phase artefact issue seems to have made a huge difference. Here are some examples of teeth I found p
772.
by Veedrac 7y ago
For the first link (game.html), normally 5-10s, but much longer or shorter for some of them. For the second (game_cropped.html), it takes much longer, like a minute or so, except when the real image contains something distinctive StyleGAN2
773.
by Veedrac 7y ago
I set up this super simple ‘Which Face Is Real?’ ( http://www.whichfaceisreal.com/ ) style challenge. Click the row to show the answers. You might need to zoom out. https://veedrac.github.io/stylegan2-real-or-
774.
by Veedrac 7y ago
> Connectionism is at heart a correlative methodology: it recognizes patterns in historical data and makes predictions accordingly, nothing more. Neural networks do not develop semantic models about their environment; they cannot reason
775.
by Veedrac 7y ago
Their compiler isn't ready and they haven't started working on an FPGA implementation yet. I personally expect they'll never launch a product.
776.
by Veedrac 7y ago
This is an unusually bad argument for him. Human intelligence being general is not the same as human intelligence being built of unspecialized pieces, or the idea that we should be able to solve all problems equally well, regardless of di
777.
by Veedrac 7y ago
> Artificial general intelligence (AGI) is the intelligence of a machine that can understand or learn any intellectual task that a human being can. That's what the term means . Anyway, this idea that human intelligence is ‘not gene
778.
by Veedrac 7y ago
JP: “As a lab, our objective is to match human intelligence.” Also JP: AGI is a “bogus concept”. ??? How can you possibly square these comments?
779.
by Veedrac 7y ago
Source? From October 2014: > The automaker says that by 2020, the factory will be shipping out 35GWh of cells and 50GWh of packs per year. https://www.zdnet.com/article/panasonic-to-pour-billions-of-...
780.
by Veedrac 7y ago
Page views have doubled since 2008, and decreased since 2015. This can't be about hosting. https://stats.wikimedia.org/EN/TablesPageViewsMonthlyCombine...

More ›