ahepp

have we tried adding "make no mistakes" to the humans' prompt?

What do you think about the prompt “The first 6 digits of pi are 3.1415”?

actual or perceived realtime / determinism requirements

TCP/IP refers to the whole protocol stack, including UDP. That is a bit counterintuitive, but well established: https://en.wikipedia.org/wiki/Internet_protocol_suite

> I don't see a reason to trust it any less than closed-hardware, closed-source solutions. I would trust a fully-open-source PIKVM more, however.

Isn't jetKVM software open source? I thought it had a public buildroot fork? Been a while since I checked, maybe something has changed.

> The accidental static bleeding from onboard hardware, coupling through satellite structures and radiating across frequencies nobody authorized.

> The numbers tell the story bluntly:

I wouldn't waste too much compute trying to parse it, lots of LLM tells in the article. Pangram puts it at 100% AI.

Doesn't the premise that there's something inferior about these AI solutions, imply that there is something superior about human intelligence and that there will continue to be some kind of useful work for humans to do?

Not the main point at all, but what was it? Pattern matching rule nestled deep in /lib/systemd/network? Some kind of mysterious netplan / NetworkManager compat layer?

You need the next token to begin the next decode. So if you can get what might be the next token in half the time, you can kick off the next decode before the true decode for this token has finished.

If the draft was wrong you can kill the speculative decode, and you haven’t lost anything except for idle time

It’s true we can’t show the user the token until we have the true decode finished, but we can launch more work internally before we’re certain

I was so optimistic about using LLMs for "write once, read many" English language documents, but the more I've used the tools, the more pessimistic I get.

More and more, I try to ask it for low prose responses because its writing just seems like such a low signal to noise ratio

I'm curious about why LLM writing fails. Particularly whether LLM writing is fundamentally flawed, or if it's just distinctive and since it often reflects low effort, that distinctive voice is associated with low quality.

I find its reliance on extremely consistent rhetorical patterns concerning. The fact that it always finds a way to talk about how "It's not the X, it's the Y Z" no matter what topic you feed it, makes me concerned that the tail is wagging the dog

>I wondered if you have tried running all of your own writings through it to verify that it thinks you are human. I was struck by Freddie deBoer's recent piece where he did this

Maybe I'm taking the "all" too literally here, but I read the article, and I'm not seeing anywhere the author ran a substantial portion of his corpus through Pangram to determine the false positive rate. That would be really interesting to see.

He does

* give an example of a piece of his writing that was, when ran in segments, flagged as generated (which he disputes)

* multiply the size of his corpus by pangram's published false positive rate and estimate that a few of his pieces would be flagged

* get the pangram model to label a piece "100% AI" when it only has 3 generated sentences

* demonstrate the ability to intentionally trigger a false positive

What kind of prompting are you using to get those results? Anything I have claude or codex write carries a ton of distinctive characteristics. Obsession with "bit-for-bit identical", "it's not the X it's the Y Z" and so on.

It's driving me nuts, I constantly have to prompt it to "explain in plain, simple English"

A bigger question I'm interested in is why do LLMs speak like that in the first place? Is that really what you get if you took the average of the English language? It would be difficult for me to believe that.

Is there something about tuning for desirable qualities that forces LLMs to have this voice?

I wonder if this is due to experience with reading technical documentation?

I think the second requires deeper concentration, but is still quite readable compared to the kind of low-content engagement / SEO stuff one read on the internet even before LLMs

The zero 2 (W) seems pretty great if you just want a cheap linux device.

I get that a lot of immutable backup solutions are code, but they don’t have to be. You can do stuff like block writes in firmware or even blow fuses on the TX traces.

>mailing list discussion (https://lists.sr.ht/~sircmpwn/sr.ht-discuss/%3CDKSTMKM0ZD9N....)

That is certainly one way to decide what to do with the data your customers paid you to store...

I think we should take the headline claim that "over 170k nonprofits lost all their data" with a huge grain of salt.

As far as I can tell, the only source for that claim is:

> And he later learned he wasn’t alone, when another company service representative called the very next day and told him that roughly 171,000 small nongovernmental organizations “lost everything” in their OneDrive accounts.

So a second hand account of remarks said over the phone by a low level employee/contractor.

That's an interesting data point to follow up on, but using a single indirectly relayed statement to make such a substantial claim doesn't seem consistent with journalistic standards.

> On Reddit, and across Microsoft’s own Tech Community forums, more such claims abound of surprise deletion without notification.

I clicked the link with text “more such claims”, and it appears to be someone complaining about being unable to reach a human after their devices were physically stolen. Not sure how that’s related to the claims in the article.

https://www.reddit.com/r/FuckMicrosoft/comments/1ub7i1h/micr...

Will that necessarily be true in the future? Would we not expect that openrouter will develop the ability to negotiate special pricing?

My understanding of it is also pretty surface level, but I was not under the impression that it develops "expertise" in a particular subject matter, at least not in a way that's easy to harness. From what I've read, it develops expertise at the token level.

Because the natural continuation of this is to say like, "Ok I want to load the bird detection expert and the navigation expert but leave the medieval European history expert behind", and my understanding is that this is not really how it works. At least at the moment.

> You can think of it like sharing.

was this meant to read "sharding"?

I understand that GitHub kinda ate the world on this, but using a branch PR as the basic unit of change seems like an arbitrary choice to me.

Torvalds invented Git for work on the Linux kernel, and the basic unit of change there is commits over email. It's not uncommon for them to take some commits but reject others.

Which isn't meant to say "pull requests are obviously wrong and commits over email are obviously correct", but rather to point out that the idea of a take-it-or-leave-it branch isn't some kind of natural unit of contribution

https://lkml.iu.edu/hypermail/linux/kernel/1402.2/00508.html

https://lkml.indiana.edu/2011.2/05416.html

https://lkml.iu.edu/2409.0/00960.html

https://lkml.iu.edu/hypermail/linux/kernel/2510.3/01369.html

I wonder if they could have added the ability to take only certain commits from a PR though?

It kinda seems like they're duplicating the "unit of change" arbitrarily, rather than just fixing the way a PR works.

I think the reason I find that a bit icky is it seems like it's diverging GitHub from the underlying git tool, which I trust a lot more.

I don't have experience with the stacked PR workflow and often find myself asking the same questions you are.

As far as I can tell, the biggest benefit of stacked PRs over just making a coherent series of commits, is that it might make it easier to start work on your second PR before you merge the first one?

With human-in-the-loop coding, that sounds like it could lead to a lot of wasted work if the first PR gets substantial feedback. But with agentic coding, I can imagine how it might be desirable to keep the agent chugging while the first PR is under review.

Interested in learning more about it and generally agree that AI is stressing the current review paradigms a lot of us are accustomed to.

Yeah my phone was doing that for a while. I just took it as a sign that maybe I should spend less time on Reddit. After a while it stopped gating me. Not sure if that was a measured decision on their end, or just that the experiment ended.

If a prefix cache is good, why not keep it in primary mem?

I got a gen10 plus microserver and liked it so much that I got a second one.

Throw FreeBSD on it and add a couple lines to /etc/exports and rc.conf and it's a NAS right out of the box

I'd probably just buy a cheap prebuilt PC with a big case, or surplus enterprise tower. You can get cheap LSI PCI cards and do miniSAS to SATA if needed.

The biggest concern I'd have with USB is power delivery to the hard drives, but I haven't even done the napkin math so maybe it's fine. The SSDs seem like they might be a waste of money. USB hard drives have a poor reputation, but I don't have a ton of experience to say how much of that is deserved. On a practical level, I'd also be concerned about knocking the cable out.

I'm guessing you mean fuel injector? The pump that pushes or pulls fuel up from your tank is not very sophisticated, afaik.

Fuel injector timing and quantity, along with ignition timing, is generally computer controlled, certainly on any modern vehicle.