Posts by matt@toot.cafe
(DIR) Post #B6NR269WJYBkGSi1J2 by matt@toot.cafe
0 likes, 0 repeats
Among other things, Miller wrote, "Linux is light weight, Solaris is a pig."It seems to me that Linux is now one of the big, primarily corporate-developed operating systems. I don't know if it's as much of a pig as 1996 Solaris was, but it has definitely put on some weight that drags it down, especially on lower-end machines.So, where do we turn for an OS that is like 1996 Linux? Maybe NetBSD?
(DIR) Post #B6QFHfX4RHwYMSGKye by matt@toot.cafe
1 likes, 0 repeats
I just read about a blind person vibe-coding a new email client for Windows. Not linking because I don't want people to pile onto this person, who is a veteran of the software industry and a respected, persistent accessibility advocate and contributor to assistive technology. Instead, I want to point out how badly the commercial software industry, particularly Microsoft in this case, has failed us such that an individual feels the need to do this. Don't know what to do instead though.
(DIR) Post #B6QFHhXUyfTSaMThvE by matt@toot.cafe
0 likes, 0 repeats
On the one hand, it's good that blind people are taking initiative to solve our own problems, rather than begging big tech companies and/or under-resourced open-source projects to give us what we need.On the other hand, I still believe that relying heavily on an LLM to generate large volumes of code is dangerous, and that we don't fully understand the pitfalls.
(DIR) Post #B6QFHjHGVzg3yatltg by matt@toot.cafe
0 likes, 0 repeats
Check out what Ron Pressler wrote about Anthropic's PR stunt of having agents build a C compiler. He looked at the actual results, not just the hype. https://news.ycombinator.com/item?id=48170244> Today's models are not yet capable enough to build non-trivial production software without close and careful human supervision, even with perfect specs and perfect tests. Without a perfect spec and a perfect human-written test suite the task is even harder.An email client falls in that latter category.
(DIR) Post #B6dKAxglYqxTTFtib2 by matt@toot.cafe
0 likes, 0 repeats
@david_chisnall The trouble is that access to smart and experienced people is very unevenly distributed; as you said, you've been very fortunate. While LLMs have been infamous for being sycophantic, my experience has been that Claude pushes back in at least some cases where that's warranted. In time, access to the best LLMs may also become very unevenly distributed, but at least for now, I think some good can come from the free or cheap access that's widely available.
(DIR) Post #B6yNvHVYzZPzOzwWyu by matt@toot.cafe
0 likes, 0 repeats
@mntmn Congrats. How quickly can it wake from that suspended state? Do you know what your power draw is in that same keyboard/USB configuration but the CPU at near idle, like, only one core powered on (if that's possible) and at its lowest clock speed?
(DIR) Post #B7MpVAnrrYgGDX5cTQ by matt@toot.cafe
0 likes, 0 repeats
@mntmn @LinaroLtd @cas Congratulations on getting this module launched! I wonder if there's room to further improve active run time. Is power management already on par with what the same SoC can achieve with, say, a Qualcomm Android BSP, or is there still more work to do in the mainline kernel? Maybe an opportunity for more funded work for the right kernel hacker?But for now, congratulations on this achievement!
(DIR) Post #B7MxpnmHSwfeWFh8W8 by matt@toot.cafe
0 likes, 0 repeats
@mntmn @LinaroLtd @cas I'm also curious if you chose the C6490P rather than the C6490 because the WiFi on the latter SoM wouldn't work with mainline Linux, or for some other reason. Thanks again!
(DIR) Post #B7Wiyqxy1pb4C0UuXI by matt@toot.cafe
0 likes, 0 repeats
@alice_i_cecile No, AccessKit. I haven't contributed anything to Bevy in quite a while, but if it has an anti-LLM policy these days and I do contribute in the future, I will of course honor that.
(DIR) Post #B7WiyrBnCQeissxx5M by matt@toot.cafe
0 likes, 0 repeats
@alice_i_cecile I started AccessKit of course, but Arnold and I are now equal partners on the project, and in fact he has been doing most of the work on it lately. We agreed years ago that neither of us would use LLMs on the project. I continue to honor that agreement, and I still think it's a good idea for the sake of keeping the project's copyright status clean.
(DIR) Post #B85tlWYtubLuz3qK12 by matt@toot.cafe
0 likes, 0 repeats
@glyph So from what you've heard and read so far, do you think there's a point where the mental damage, or whatever you choose to call it, from usage of LLMs is irreversible? If one hasn't yet reached that point, does merely stopping usage of LLMs reverse the damage, or does one have to do something more to restore what they had?
(DIR) Post #B86O7qVBS9qzp3fEnY by matt@toot.cafe
0 likes, 0 repeats
@mntmn I'll be curious to find out if typst's HTML output is working well yet. But even if you can only (or only want to) generate a PDF version, typst will at least generate a tagged PDF document, so it should be accessible with a screen reader.
(DIR) Post #B8czdSOlcnBWrfcJQe by matt@toot.cafe
0 likes, 0 repeats
@ariadne I wonder how common this problem is in Rust, and if there's anything about Rust that makes it hard to avoid excessive mallocs. Anecdotally, it seems fairly common to use an alternative malloc in Rust projects running on musl.
(DIR) Post #B8fjz2m9JYPLGeteim by matt@toot.cafe
0 likes, 0 repeats
@ariadne As far as I'm concerned, if OpenAI used anything less than Xen via Edera for their sandbox, they were asking to have the agent escape.
(DIR) Post #B90FPgz0yWmTKOJADA by matt@toot.cafe
0 likes, 0 repeats
@ariadne Is that a bad thing?
(DIR) Post #B90IXezsXj6vLbCfBY by matt@toot.cafe
0 likes, 0 repeats
@ariadne Or did I misunderstand? I read "what the fuck" as a negative reaction.
(DIR) Post #BAM0GlvWy6rivgeddo by matt@toot.cafe
0 likes, 0 repeats
Some of the takes here on fedi are just unhinged. Someone I follow recently boosted a post that said "fuck the Rust Foundation", because the Rust Foundation is a 501(c)(6), which apparently allows "the worst organizations on the planet" to be members. I can't bring myself to agree with such a take, because I think Rust is really important and unique, and regardless of what else these organizations are doing, I'm glad they're funding it.
(DIR) Post #BAM0Gm9i7OCxdfHxk8 by matt@toot.cafe
0 likes, 1 repeats
If I'm honest, part of my reaction is that I hate the thought of possibly having to give up Rust's combination of safety *and* efficiency if Rust comes to have the wrong political associations.
(DIR) Post #BAM0GoWpIPt2zw7S3E by matt@toot.cafe
0 likes, 0 repeats
I suppose some might say I have my priorities wrong, that social justice and solidarity outweigh technical superiority. But for me, reliable and efficient software is (largely) a matter of caring about the people that use that software. (Largely because, yes, small and fast software is also about the kind of aesthetic that only matters to computer nerds.)
(DIR) Post #BAWhaOPyzBEM8zr5Lk by matt@toot.cafe
0 likes, 1 repeats
A follow-up on the axum-marko project that I posted about a few weeks ago, in the parent post: I now consider this hastily done slop, and I regret having published it. I've deleted it from GitHub and crates.io. This doesn't mean I renounce all use of LLMs for coding, but this project was precisely the kind of foundational software that deserves to be implemented carefully, not hastily and without thorough review. So I've withdrawn my earlier addition to the slop in the world.