angry_octet

5 ms·

1.
by angry_octet 4d ago
BTW you can swap over NFS and iSCSI with swap files, with NFSoIB, or use the Infiniswap modules: https://github.com/SymbioticLab/Infiniswap Great hack for when you have a massive working set but don't want to rewr
2.
by angry_octet 4d ago
The swap files do use a non+VFS path after setup in 2.6+ kernels, but it is complex and has had bugs: https://lkml.iu.edu/hypermail/linux/kernel/2103.0/06524.html The bug is in here: https://g
3.
by angry_octet 4d ago
As @mitxela says, if you need to fault in a page which is needed for some critical OS function, which requires going through the VFS layer, and that requires paging in another page, which has been swapped out due to memory pressure ... Ther
4.
by angry_octet 5d ago
It is obvious to me why rubygems was used, when the data was public: their firewall had whitelisted connections to package repositories like rubygems, npm, PyPI, etc, so they used those sites to proxy to their destination.
5.
by angry_octet 5d ago
High performance databases still use raw I/O. You get much better control of data placement and write ordering, without filesystem confounders. With traditional spinning rust being replaced with SSDs and SMR it can be even more importa
6.
by angry_octet 5d ago
It's actually incredibly sensible not to just make changes to core OS settings unless you know what you're doing. There are good reasons not to use swap files, which become evident in OOM situations.
7.
by angry_octet 5d ago
This is such an incredibly annoying and deliberate defect. I want an extension that will let me resolve the true url without visiting the site, or even rewrites the entire page to show the url.
8.
by angry_octet 6d ago
The only ethical path for OpenAI was to offer infinite free credits and tooling support. Trying to gazump them is reprehensible.
9.
by angry_octet 6d ago
People seem very angry about my comment, which perhaps indicates why WordPress is in such chaos. The SEC has significant enforcement powers still, and investors can mount lawsuits against shareholder oppression. https://www.dando
10.
by angry_octet 7d ago
I should have said: establish sole ownership.
11.
by angry_octet 7d ago
Regardless of who he puts on the board, directors have legal obligations to the SEC and all shareholders. If he wants that level of control he would need to take the company private, and various investor agreements probably prevent this.
12.
by angry_octet 7d ago
I don't think your reading is correct. ML history is related with examples of models using hints from side channels. If it would be possible to replicate OpenAIs success with a model cut-off earlier than the rumours, and without hintin
13.
by angry_octet 8d ago
You are entitled to your opinion, but it isn't supported by the information already available.
14.
by angry_octet 8d ago
Everyone?? No, most definitely OpenAI. But they have learnt their lesson, next time they won't reach out to who they stole it from, they will publish first.
15.
by angry_octet 8d ago
OpenAI cribbing from other researchers. We just have to assume OpenAI is actively adversarial in future. Accidental cyber intrusion is also well within model capability.
16.
by angry_octet 9d ago
I think Shakespeare would love TL;DR, but I think the pause in 'tee -- ell -- dee arr' will go in favour of vowels, leaving 'tillder' or 'tellder'.
17.
by angry_octet 20d ago
The Iran war and SoH crisis has been great for EV uptake outside of the US. In combination with the US AI apartheid, it has been a huge boost for Chinese EVs and Chinese AI adoption.
18.
by angry_octet 20d ago
I think its nothing like the Enterprise version, which is why they hate having to maintain it, same with Atlassian and Azure/Entra AD. Once service to rule them all, and in the cloud bind them. For security reasons, I would love a low
19.
by angry_octet 20d ago
It is quite possible for github load balancers to route paying customers to different queues/infrastructure. I think much of the bot activity is paying customers, they just don't have per-client QoS because that would be hard. And
20.
by angry_octet 20d ago
We are. But we are also experienced at dealing with temperature extremes. The problem is that the 45C temperatures we regularly experience, and Europe just had for the first time, are going to be 50C and on top of prolonged drought and floo
21.
by angry_octet 20d ago
It's the garden variety violence in the States that gets you.
22.
by angry_octet 21d ago
Unfortunately it is not so simple. Once you provide a source of intelligence that is accessible over the network you are supercharging any software that can access it. Web interfaces (chat) designed for humans can very easily be used by pro
23.
by angry_octet 21d ago
This is misunderstanding the attack. The potentially malicious output tokens are processed entirely by classical text processing techniques, i.e. parsing, in order to enable tool calling on the client and via MCP servers. Garden variety v
24.
by angry_octet 22d ago
You get to the engine via http. Also once you've exploited the engine instance / host you can C2 via http. We're not talking about vulnerabilities in http. There is lots of surface inside the engine, see links in https:&#x2F
25.
by angry_octet 22d ago
Different models test to find different attack paths. Attackers with bigger libraries of attack techniques will be able to train more dangerous models. There will also be models that make better use of tools, like static analysis and fuzzin
26.
by angry_octet 22d ago
The attacker supplies the context that triggers these bugs via the http interface. There is no other way to supply it with context. Vulnerable code could exist in token generation and in hook recognition and tool call parsing [1]. There is
27.
by angry_octet 22d ago
@hobofan How would you access that parser?
28.
by angry_octet 22d ago
How do you think the tokens get to the vLLM instance? Telepathy?
29.
by angry_octet 23d ago
Anthropic's watermarking technique is possible, and worryingly could effect low frequency choices (eg as I said, with object placement or style). However, this has to be a generation time decision, you can't really do that with a
30.
by angry_octet 23d ago
There are side channels available, they have licenced IP blocks, not a whole CPU. All that logic for the cache and interconnect.

More ›