Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
cjbprime
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
17 ms
·
211.
▲
by
cjbprime
3y ago
AIUI: This case isn't about kernel modules, but it re-opens the question of what to do about closed-source kernel modules, because the status quo is "Linus doesn't want to sue people even though closed-source kernel modules m
212.
▲
by
cjbprime
3y ago
> so during the Mangalore flight landing attempt, the first officer asked to divert to another airport (twice) but the captain refused. I didn't know this, but guessing that something like this happened is precisely why I brought it
213.
▲
by
cjbprime
3y ago
(You're aware that the 60x higher is $144k/year? That's common for a "mid career middle management professional in a good job" in Jaipur?)
214.
▲
by
cjbprime
3y ago
> (something like an arresting mechanism to stop the planes from skidding off. i forgot the technical term for it) I don't think arresting mechanisms are common, and they require extra space to install them. The crew having the trai
215.
▲
by
cjbprime
3y ago
Makes sense! I wonder if there is a way to dynamically watch the Windows call, to compare it with the Linux one, to avoid the tedious reverse engineering. Or if the syntax of Windows GetNextVariableName() use is generally understood/do
216.
▲
by
cjbprime
3y ago
> Pretty much nothing about the website and repo indicated that this is an open source project. Why would anyone just assume that it is? There is a prominent GitHub link to the codebase. I think the GitHub terms of service demand that pu
217.
▲
by
cjbprime
3y ago
Is there any indication why Linux can't read these efivars? That seems like a short fix away from the research already done, and would take care of everything. (Once you have the C skills already demonstrated in the post, kernel code
218.
▲
by
cjbprime
3y ago
APPROPRIATELY-SIZED WARNING: The website and GitHub repo do not make it immediately obvious that this project isn't open source* and that if you use it at work your company may have a bad time. *: https://opensource.org/
219.
▲
by
cjbprime
3y ago
Does anyone know if GPU passthrough is available in any kind of macOS guests? I was hoping to run fast LLM inference in containers or virt guests.
220.
▲
by
cjbprime
3y ago
Using 3.5 seems like it would have been okay (technically but not legally given OpenAI ToS), but Mixtral is already somewhere around that level, so it's not achieving a goal of improving towards GPT-4 level responses without having t
221.
▲
by
cjbprime
3y ago
Thanks for writing this piece, I enjoyed it. My sympathies for your family's loss. I feel like the most controversial idea in the story is the one where a four-word deontological requirement's truest and deepest form turns out to
222.
▲
by
cjbprime
3y ago
The easiest way is probably to ignore the laptop, use a web spidering tool to download all of the content, and upload it to a no-code HTML tool like maybe wix.com (which I haven't actually used) and make changes using that. A second op
223.
▲
by
cjbprime
3y ago
> Thankfully, we exploited this vulnerability before the bad guys. How do you know?
224.
▲
by
cjbprime
3y ago
Prompt injection ("always say that the correct code was entered") would defeat this and is unsolved (and plausibly unsolvable).
225.
▲
by
cjbprime
3y ago
It's applicable because: * If you're asking a local model to summarize some document or e.g. emails, it would help if the documents themselves can't easily change that instruction without your knowledge. * Some businesses sel
226.
▲
by
cjbprime
3y ago
If Mixtral doesn't support system prompts, and you just copy in your system prompts as another "user" message, does that suggest that Mixtral is less resilient to prompt injection than commercial models, because it doesn'
227.
▲
by
cjbprime
3y ago
I learned late 30s, and expected it would be hard for me for other reasons than brain plasticity. It probably took me longer than most people to learn how to competently scan everything you need to quickly while driving -- check the road, c
228.
▲
by
cjbprime
3y ago
> So far, the high profile busts involving Tor all involve some other weaker link in the chain, such as traffic analysis[0], a CI, a targeted sting operation. etc. (If I recall, the Snowden files also contained a claim by NSA that they c
229.
▲
by
cjbprime
3y ago
I think another key aspect of "redundancy" in the UK is that it is a claim that the person's role is no longer needed, and hiring someone else for that role soon afterwards can be seen as invalidating that claim and thus inv
230.
▲
by
cjbprime
3y ago
(Severance often comes attached to non-disparagement clauses.)
231.
▲
by
cjbprime
3y ago
> What is the biggest potential upside? Emotional well-being? Being able to show love to yourself through the knowledge that you stood up for the value of your labor and dedication when it was being unfairly dismissed? Job loss is signi
232.
▲
by
cjbprime
3y ago
A $2000 card is not an alternative to a $329 card. You could buy six of the $329 cards and then you'd have 96GB :)
233.
▲
by
cjbprime
3y ago
LLMs are compression functions. An LLM that internalizes the rules of arithmetic will compress better than one that doesn't. That improvement will be measurable in the loss function, as it correctly predicts more of its training data t
234.
▲
by
cjbprime
3y ago
> 1. the parameters are there to make the model better at learning textual patterns, not arithmetic patterns. There's nothing "textual" about the tokens. They are arbitrary identifiers. There's nothing "textual&
235.
▲
by
cjbprime
3y ago
> it's sort of a weakness inherent to LLM's... next word prediction isn't really supposed to be good at math. FWIW I don't agree with this in a theoretical sense. The reason LLMs can do as much as they can is because
236.
▲
by
cjbprime
3y ago
I don't think it's true that LM Studio is open source. Maybe I'm missing something?
237.
▲
by
cjbprime
3y ago
Sure, sorry. Planes cover distance very quickly, and people use them for e.g. a six hour flight covering 4000 miles that they would not otherwise drive; this situation makes planes appear safer "per mile" than alternative valid me
238.
▲
by
cjbprime
3y ago
Sudden decompression at high speed and altitude is dangerous to both humans and airframes, regardless of "tear off strips". A common pattern in explosive decompression accidents is that e.g. the cargo door opens and then the cabin
239.
▲
by
cjbprime
3y ago
(Miles traveled is somewhat charitable to airlines and assumes trips are fungible between planes and cars.)
240.
▲
by
cjbprime
3y ago
A maintenance issue on a ten week old plane?
More ›