Veedrac

55 ms·

721.
by Veedrac 7y ago
> Compare that to remote backdoor access to everyone's encrypted communications. All they have to do is push a button in order to instantly get access to someone's mail, calls, instant messages, social media, phone location his
722.
by Veedrac 7y ago
It's not just that it's easy, but that the theft is legal and profitable. You don't even need to be rich to do this, you just need stuff you can claim; see How to Break YouTube (Copyright Claim your own video). https:/&
723.
by Veedrac 7y ago
In which a science fiction writer weaves a fantastical tale about how the Doomsayers are making their decisions from fantastical tales and emotions, instead of actually reading and responding to the arguments they present. The irony is so p
724.
by Veedrac 7y ago
Note that Google's system operates in pre-mapped areas and uses lidar.
725.
by Veedrac 7y ago
https://www.reddit.com/r/Neuralink/comments/ce8a5i/new_neura...
726.
by Veedrac 7y ago
Conversation doesn't work when people immediately strawman the other side.
727.
by Veedrac 7y ago
One of them has to give. You can't effectively police what's perfectly anonymous.
728.
by Veedrac 7y ago
Starlink will sell at pretty much whatever price it needs to, since serving data has effectively no marginal cost. The only unavoidable one is the ($300) base station, but people can share.
729.
by Veedrac 7y ago
This absolutely doesn't explain C++'s weirdness. Those template parameters don't make anything faster, its broken seeding ends up pretty slow, and a Mersenne Twister is hardly performance- (or anything-)optimal. Heck, some of
730.
by Veedrac 7y ago
C++ suffers this a lot particularly because its paper system rewards pages of meaning-free implementation details with little to no qualitative discussion. Where else can you find a random library that lets you template a Mersenne Twister w
731.
by Veedrac 7y ago
I'm not actually sure, but DeepL[1] is probably the best online one, and Google's larger M4 models[2] are probably up there too. [1] https://www.deepl.com/translator [2] https://ai.googleblog.com/2
732.
by Veedrac 7y ago
I don't disagree that translation is far from mastered, but bear in mind that Google Translate isn't state of the art, mostly because of computational constraints, and 2016 GTranslate was even worse.
733.
by Veedrac 7y ago
MuZero is in some sense the proto-holy grail, in that it implements learning and planning into unstructured tasks over purely internal models. While there is an obvious chasm between it and the end point, this is still something that has on
734.
by Veedrac 7y ago
> Cutting-edge research does feel like it's reaching a plateau It's really not. The second half of last year alone had MuZero and Megatron-LM, to name just a couple that most scream to me that we are actually progressing toward
735.
by Veedrac 7y ago
Consciousness is empirical, since you can talk about it. Therefore if panpsychism is not empirical, it's wrong.
736.
by Veedrac 7y ago
At what point in development does a child become capable of hearing and interpreting sounds? Can a human hear sounds even if the cells in it cannot? Do ant colonies contain a primitive hearing because they react to vibrations in the air? Do
737.
by Veedrac 7y ago
No, it holds much more strongly than that. The idea that consciousness can only arise in neurons shaped exactly like human neurons—and not, say, slightly larger neurons like exist in many other animals, or neurons with slightly different en
738.
by Veedrac 7y ago
Panpsychism is often claimed to be a solution to these things. I've never actually seen any argument that would defend the thesis. I say panpsychism is obfuscated dualism because it involves no causal mechanisms anywhere, and even wo
739.
by Veedrac 7y ago
You're right, but only in the sense that panpsychism doesn't answer any questions. Literally any. The only thing it does is split the question of ‘where does human consciousness come from?’ into ‘how does human consciousness reduc
740.
by Veedrac 7y ago
Panpsychism is incredibly silly. It's just obfuscated dualism. It literally doesn't explain anything , and it's obviously wrong for the simple reason that it gives no way to explain why humans act on their conscious experi
741.
by Veedrac 7y ago
Alice: We can agree the brain is conscious, yes? Bob: Of course. Alice: But what if you scaled up all the neurons in the brain a factor of 2? Would it still be conscious? Bob: Sure, if it's the same system running the same way. There&#
742.
by Veedrac 7y ago
The i432 was well before my time, so it's hard to account for the privilege of hindsight, but I found myself looking in horror at most of its design. Hardware needs to account for the fact it's ultimately a piece of physics, and t
743.
by Veedrac 7y ago
Sure, if everything prior to actively mass manufacturing counts as R&D, but they've proven they can manufacture, that the product works and outperforms the competitors, and they have customers. I was objecting to ‘early stages star
744.
by Veedrac 7y ago
They've shown excellent pre-production models, and have $300m in ‘confirmed commitments’. I'm not going to bet they'll be on time, but it seems wrong to dismiss them as just talk. https://www.youtube.com/watch
745.
by Veedrac 7y ago
> I'm still waiting with bated breath for a reliable e-ink computer monitor comes out with a decent refresh rate. I'd even be happy with just greyscale. CLEARink have a best-of-all-worlds high-refresh rate color e-paper that wa
746.
by Veedrac 7y ago
Could you be more explicit? What about the naïve approach to training (same graph but backwards, computing gradients) is going to fail? Wrt. matmul, if you couldn't split them up, today's AI accelerators wouldn't work full s
747.
by Veedrac 7y ago
GPT-2 uses attention, which is very memory hungry to train, so probably won't work well. But I agree with your overall point.
748.
by Veedrac 7y ago
I would be surprised if Cerebras was trying to handle any recurrence inside the overall forward/backward passes. It seems like a lot of difficulty (as mentioned) for peanuts. I don't get your point about training. Yes, it's b
749.
by Veedrac 7y ago
No, you don't have all the issues with traditional NUMA because you aren't doing the same sort of heterogeneous workloads. You're always working on local data, and streaming your outputs to the next layer. This isn't a r
750.
by Veedrac 7y ago
Remember that you aren't compiling arbitrary programs. Neural nets don't really have any local looping control flow, in the sense that data goes in one end and comes out the other. You'll have large-scale loops over the whole

More ›