radlad

6 ms·

1.
by radlad 17d ago
I got this wrong yesterday. While LUKS2 and GRUB now support Argon2, I configured LUKS1, which uses PBKDF2-SHA512. In any case, it is GRUB which behaves slowly during boot. While I haven't tested Argon2, I believe it is affected in the
2.
by radlad 17d ago
I'm reminded of LUKS Argon2 implementation, which is much slower than userspace. Therefore a sufficiently difficult challenge that may take less than a second in userspace Linux can take 20 or more seconds at boot.
3.
by radlad 22d ago
You can create a custom filter for this kind of thing, fwiw.
4.
by radlad 24d ago
I have an MSI OLED monitor, and if I accidentally turn it off (trying to find the input switcher), and then back on, it shows this warning. But it automatically runs when it's turned off. The warning is shown because I interrupted it r
5.
by radlad 25d ago
On censored inputs only.
6.
by radlad 25d ago
> What makes this build different is the word before FP8: uncensored. We applied abliteration — orthogonalizing the refusal direction out of the residual stream — to remove the model's safety-alignment refusals. The result is a mode
7.
by radlad 27d ago
> You need context to really understand a book or a work of art. You want to know what kind of environment produced it, what it's talking to. I agree with this premise - but the arts and humanities are a big part of how we learn the
8.
by radlad 29d ago
Oh wow, you might've just explained some symptoms I've run into a few times. Thanks for your comment.
9.
by radlad 1mo ago
There are definitely different tiers. I've bought simple shelving that began to fall apart within years (Billy maybe?). I have 10 years old bedroom furniture still going strong (Kallax/Malm).
10.
by radlad 1mo ago
Until now, human knowledge and values have built on prior human knowledge and experience. If AI is able to develop without human influence, I believe its value system will necessarily diverge into something alien.
11.
by radlad 1mo ago
I have long suspected but not confirmed that /btw uses a lower tier model like Haiku. Depending on how you're triggering reviews, you may be using a sub-agent?
12.
by radlad 1mo ago
Recently learned I can just add a question mark to the end of my Kagi search to get an assistant answer. Kagi was already great at surfacing the most relevant pages, but now I often don't even need to click.
13.
by radlad 1mo ago
It's possible reflected instances are masked too, like GitHub Actions. But I don't know.
14.
by radlad 1mo ago
Love Incus and I'm using throwaway restricted projects for testing. Highly recommend incus-windows if you need to do any Windows testing. Having agents validate Windows behavior has reduced so much toil for me.
15.
by radlad 1mo ago
Not cherry-picking, I quoted the entire first paragraph of the section titled "Definition and purpose." > LLMs will can not and will not ever do these things Once again, I am not claiming LLMs would replace an OS . This is a
16.
by radlad 1mo ago
Are we reading the same Wikipedia page? > An operating system is difficult to define,[7] but has been called "the layer of software that manages a computer's resources for its users and their applications".[8] Operating sy
17.
by radlad 1mo ago
"The definition"? Whose? There are many definitions of "operating system" and I think it may be you who has it wrong: https://en.wikipedia.org/wiki/Operating_system#Definition_an... Having a kernel,
18.
by radlad 1mo ago
They load only when operating on files in that directory. i.e. Lazy loading? For example, I have one in my schema migrations directory that explains how to write migrations with appropriate concern for forwards- and backwards-compatibility
19.
by radlad 1mo ago
I'm obviously not saying that LLMs replace the glue between hardware and software. That would be absurd. An operating system is just the interface a user has to the programs they run on their computer. Surely you can imagine a world wh
20.
by radlad 1mo ago
I'm not advocating for this model. But I'm also not sure it has any more lock-in than choosing an OS today: Windows, macOS, Linux, ChromeOS, Android, iOS.
21.
by radlad 1mo ago
I already mentioned this: > This obviously isn't entirely true: [...] it's the way I access my LLM. It doesn't change my opinion that there's a very real possibility that we move to thin clients or microkernels intend
22.
by radlad 1mo ago
> I don't get asked about access for Chrome at every update. Is it possible that you clicked "Approve" while the parent clicked "Not now"? One of my personal issues is that I was using software called "Secre
23.
by radlad 1mo ago
I think because the idea of an "agent operating system" is starting to feel more and more inevitable. Since I can use AI to: - Pull information from external sources - Transform and/or augment that information - Publi
24.
by radlad 1mo ago
That's what MEMORY.md and subdirectory-specific CLAUDE.md does.
25.
by radlad 1mo ago
Yes, my question is directed around how you are determining "task result." Is this measuring whether the code works, or whether it is maintainable? As with human-powered coding, we read code far more frequently than we write it. I
26.
by radlad 1mo ago
'better results' in terms of what though? A benchmark, or code that I would actually click "approve" on in a pull request scenario?
27.
by radlad 2mo ago
I find it impossible to daydream while focused on traffic.
28.
by radlad 2mo ago
This is blatantly false based on my usage of Fable and Opus. Though they are much better than they used to be.
29.
by radlad 2mo ago
As a native English speaker, I worry about picking up LLM-speak simply due to interacting with Claude Code all day.
30.
by radlad 2mo ago
I'm sorry to hear that's how it's going at your work.

More ›