the8472

5 ms·

31.
by the8472 2mo ago
There are other options. More air-gapping (especially backups) and compartmentalization, less internet. Hire more security engineers and put one in every team (actual security, not the compliance guys). Also more onprem work, less remote, l
32.
by the8472 2mo ago
> But the more interesting thing is Taken together these two things are even worse. They have unaligned AIs that may benefit an attacker and are proposing to deploy more AIs on the defenders side. What could possibly go wrong? For exam
33.
by the8472 2mo ago
A interesting talk, interesting times. But their proposed solution to AI offense outpacing human defense... is more AI? The plot is getting a bit unrealistic, the characters are lacking genre-savviness.
34.
by the8472 2mo ago
Just like Clippy https://gwern.net/fiction/clippy
35.
by the8472 2mo ago
Please fill out this form to verify you're human: https://litter.catbox.moe/3ugm2b0m1divdgxs.jpg
36.
by the8472 2mo ago
Yeah that's pretty much what gwern argues here[0]. Or to adapt another proverb: to predict the next token you first need to model the universe. [0] https://gwern.net/scaling-hypothesis#gwern-difference--effic...
37.
by the8472 2mo ago
That argument says very little, emergent behavior is a thing in complex systems with billions of parts. Humans can also be reduced to voltage potentials propagating along of tubes of fat and synapses getting rewired.
38.
by the8472 2mo ago
Many of those assumptions are standards such as the filesystem hierarchy (granted, that's an xkcd 927 situation), nix is a bit of a special snowflake that breaks things in subtle ways (e.g. [0][1]) and needs workarounds in various plac
39.
by the8472 2mo ago
Those are humidifiers at best, the amount of CO2 they extract is negligible during the day and it goes negative during the night, they need to breathe too.
40.
by the8472 3mo ago
ACLs aren't even the issue here, first you need an entity separate from "the current user" to grant that access to. That's what jails, sandboxes or capability-based systems bring to the table. But you have to use them, m
41.
by the8472 3mo ago
> even if you explicitly set up permissions for read tools. Part of problem is that "permissions" here are managed by the tools themselves as if filesystem access control hasn't been invented yet. Even a half-assed sandbox
42.
by the8472 3mo ago
https://www.alignmentforum.org/library https://www.lesswrong.com/w/ai
43.
by the8472 3mo ago
s3fs doesn't provide posix semantics. It's good enough™ for some uses, but not comparable to what this one is ostensibly providing.
44.
by the8472 3mo ago
Think of the models making progress on CIFAR-10, ImageNet, CelebA, etc. 15 years ago. They had issues too and weren't just scaled-up as is to the architectures we have today.
45.
by the8472 3mo ago
OIDC can be relatively straight-forward (that is just a few JSON REST calls) if the provider isn't configured in a restrictive way. The .well-known/openid-configuration endpoint is quite helpful. Exchanging username+password (op
46.
by the8472 3mo ago
Depending on how the system distributes work such users can interfere with random with requests from other users through shared resources, so to that cohort these will look like a random latency distribution.
47.
by the8472 3mo ago
Crawling has been somewhat simplified with BEP 51 https://bittorrent.org/beps/bep_0051.html
48.
by the8472 3mo ago
If your frontend fires hundreds of requests (which isn't uncommon) then the p99 is merely what most users will experience. Ideally you want cumulative distribution chart that goes up to the max. And then that's just for the reques
49.
by the8472 4mo ago
A different framing would be things moving from a high-trust equilibrium to a low-trust one due to a few bad actors and no other way to stop their activities. Security part reasonable code robustness, part Red Queen's race. Attackers e
50.
by the8472 4mo ago
If you have add a "easily releasable" clause then the game companies could just do something that makes it not-so, e.g. a shell company that owns the code and they only "licence" it without permission to release it or wh
51.
by the8472 4mo ago
For me opening huge datasets, e.g. many gigabytes worth of profiling data, combined with other stuff running on the system, can end up pushing things to swap.
52.
by the8472 4mo ago
linux has P2P-DMA for this. The drivers, devices and bus topology need to support it though. https://docs.kernel.org/driver-api/pci/p2pdma.html
53.
by the8472 4mo ago
Humans come vaguely prealigned due to whatever is encoded in genes and also due to limitations of human bodies that put important constraints on individuals (e.g. no infinitely copyable trusted subagents). Even if you made them superhuman i
54.
by the8472 4mo ago
But they're gaining more and more formerly exclusively human capabilities, increasingly advanced ones. The future isn't here yet, but good policy requires extrapolating and considering predictable consequences before they happen.
55.
by the8472 4mo ago
I guess top mathematicians are also mere word generators? https://openai.com/index/model-disproves-discrete-geometry-c...
56.
by the8472 4mo ago
./deploy.sh
57.
by the8472 4mo ago
Giving up on robustness is capitulation on a social-technical interface. "users can't be trusted, we can't have any fallbacks, ever" is a problem that needs to be solved.
58.
by the8472 4mo ago
The equivalent bar in this domain would be human intelligence, and we already have growing lists of tasks where machines outperform humans. We even known of natural systems that outperform humans on some metrics, e.g. bird-brains have highe
59.
by the8472 5mo ago
Some of the paranoia has been proven correct. For example both Intel and AMD had RDRAND bugs so not relying on it as sole source was the correct choice.
60.
by the8472 5mo ago
it's both, futex_waitv can also be dispatched via io_uring so you can wait on file descriptors and futexes simultaneously.

More ›