teravor

6 ms·

1.
by teravor 5d ago
if you compare an embedding model to something like Jev which asks 100 questions and use the answers as the embedding you will be able to get move mileage out of the latter. especially because you don't need to train any classifiers fo
2.
by teravor 6d ago
> Just keep growing and pay a fine if regulators ever find out. it's unfortunate that they weren't able to get a lawyer with experience to comment on whether or not he is right. would certainly be interesting to know
3.
by teravor 6d ago
memory encryption is cheap. securing the pathway isn't particularly difficult (it's probably decoupled from the TEE monolith) for example every TPU/GPU has its own private key and the devs load the weights into it by sending
4.
by teravor 6d ago
the tokens are generated by hardware with secure enclaves (encrypted weights) and then sent over a network to some remote CPU where they can manifest an effect. it's not much different during training. how exactly are they supposed to
5.
by teravor 8d ago
some password managers can pretend to be a passkey. but I don't use them because it requires finicky brittle configuration (and docs lookup whenever you want to redo it). if passkeys had a simple text challenge response text option it
6.
by teravor 9d ago
no, it's "buy lottery tickets to exclusive lotteries". the ability to enter such lotteries is where success lies. either through connections or ability.
7.
by teravor 9d ago
> immediate redemption with the issuer once it's been spent/accepted. This is a terrible model for physical cash not when everything is now online. also the obvious way such cash would work is that "redemption&q
8.
by teravor 9d ago
it would depend on the government and the situation. I suspect most don't realize this is possible. a small government can generate immediate demand for their currency doing this.
9.
by teravor 9d ago
when you unblind a blind signature all the signer knows is that it's a signature they signed at some point, they know nothing (true zero knowledge) about when or where they signed it among all the other signatures.
10.
by teravor 9d ago
combine with a central authority server network and make it blind signatures and you have untraceable e-cash.
11.
by teravor 11d ago
you would probably need special engine optimizations to really make use of it, and VR hasn't been exactly an attractive option for AAA games.
12.
by teravor 11d ago
I suspect the answer is actually very simple, when you compare this device to one that is made deliberately dumber there are far too few actual differences to make it make sense.
13.
by teravor 12d ago
usually, the implication of ZKP is that you buy coupons and claim them without attribution. in this coupon scenario the ZKP can just be a blind signature scheme. however signal has an obscene fondness for TEEs (secure enclaves) so they may
14.
by teravor 12d ago
> inference is already being subsidized idk where this comes from but it's laughably false. the only place where actual subsidization (below cost) might be happening are the subscriptions. even that is unlikely because to
15.
by teravor 13d ago
with Wayland it's generally not any safer. if you use Wayland's security context to prohibit privileged protocols such as arbitrary clipboard access then an application will either not be able to grab clipboard content until you f
16.
by teravor 13d ago
when an LLM is trained backpropagation reaches into the entirety of the key value table. or rather, all the layers (not just the tokens) which come before the current tick. langauge is only produced by the final layer every tick. and every
17.
by teravor 14d ago
> They're still autocomplete LLMs are simulations and the tokens are the ticks. if we transcribe your brain into a simulation and give it a tickrate, you will be just autocomplete too. the argument could be made that you a
18.
by teravor 14d ago
> Many graduate students (I know) are having a crisis if any of their research worth it? If AI can (or will) do everything, what's the point of doing experiments and all? This will eventually deter a whole generation of curio
19.
by teravor 15d ago
it's only faster in a medium, which technically isn't even light but rather electron interactions.
20.
by teravor 15d ago
apparently they weren't even properly configured until almost touching down.
21.
by teravor 15d ago
if such a method gains traction then so will the effort to bypass it. either by stealing private keys or just projecting light onto the sensor.
22.
by teravor 16d ago
some VMMs just straight up translate vsock from the guest to UDS on the host.
23.
by teravor 18d ago
> “you are saying I am doing X bad thing? What about Y bad thing you’re doing, let’s discuss that instead”. no one is generally expecting or even asking for "let’s discuss that instead". what they are doing is trying
24.
by teravor 19d ago
false security. if there aren't LLM's monitoring patches with disassembler MCPs ready, there will be soon.
25.
by teravor 21d ago
well that's why setting it up is hard, because you would want to do it in a way that what they want doesn't matter.
26.
by teravor 21d ago
ideally, an auction and the vendor or a government can bid against malicious actors (which can also be a government). hard to set up though.
27.
by teravor 21d ago
RCE inside sandbox, so requires chaining with another 0day.
28.
by teravor 22d ago
there have been people who took existing LLM's and conducted an algoritmic search to find out which group of layers they can duplicate in order to improve performance, and it worked.
29.
by teravor 22d ago
the only real way to have a decentralized domain system is something like https://github.com/pubky/pkarr but you lose the ability to have short domains. also until such a time that pkarr is widely adopted, you are bett
30.
by teravor 23d ago
> Researchers have confirmed Graphite attacks through WhatsApp and iMessage, although the complete method remains secret. unsurprising that it's closed source software which is vulnerable like this. LLMs can now decompile

More ›