pugio

Cut & Paste always struck me as more of a Windows thing. Which is weird because obviously I can do cmd+X / cmd+V on a mac... I just never do. Instead it's "select" - command+C - delete - cmd+V.

Am I the only one who does that?

(EDIT: Same thing on Finder. cmd+C cmd+Del cmd+V)

Anyway, I love the design aesthetic of the site, and when people question existing paradigms. Will definitely be checking Ishmael out.

Exactly the same - took way more work to delete the thing.

Yes, that is precisely what happened. Over and over again. Every time I had to delete it. It was something to do with the way they embedded Google login on the page.

This was worse than that. This was no click. It would literally do it without any intervention as long as I browsed to the page.

LinkedIn is one of the most annoying sites on the internet. Every time I mistakenly click a link, it automatically grabs my Google account and creates a new account and profile for me, which I then have to go in and delete. Yes, I finally figured out which arcane preferences settings across both LinkedIn and Google I had to tweak to stop this from happening, but I think it was only after ChatGPT was around to help me wade through the mess.

References:

Information hazard: https://en.wikipedia.org/wiki/Information_hazard

Babel: https://en.wikipedia.org/wiki/Tower_of_Babel

On the Calculation of Volume: https://en.wikipedia.org/wiki/On_the_Calculation_of_Volume

The revenge of NLP

The tool uses good old-fashioned regex/NLP—no LLMs—and rewrites the visible LLM Markdown without touching files on disk. I made it by analysing 10M+ words of past pi transcripts, doing frequency and pattern analysis, and iterating on rules to strip out the most egregious sloperisms. Now the circle is complete.

BEFORE

"Honestly, you're absolutely right—and that's a great question. Let me be clear: the fact that the parser works at all is actually quite remarkable. We utilize the tokenizer in order to handle the majority of edge cases, and due to the fact that it's recursive, it processes a large number of inputs. It's worth noting that this isn't really a bug—it's genuinely a deliberate design choice. In terms of raw throughput, the cache is significantly faster, and I think we should simply ship it.

That said, at the end of the day, not only is the design correct, but it is also fast. In other words, prior to refactoring, we should clearly delve into the nuanced, robust architecture. In conclusion, this is a testament to the seamless, intricate work the team has done."

AFTER

"That's a great question. That the parser works at all is remarkable. We use the tokenizer to handle most edge cases, and because it's recursive, it processes many inputs. This is a deliberate design choice. For raw throughput, the cache is faster, and we should ship it.

Not only is the design correct, but it is also fast. Before refactoring, we should delve into the nuanced, robust architecture. This is a testament to the seamless, intricate work the team has done."

I wrote a thing about exactly this, but I'm resistant to blogging for undefined reasons so, maybe this will help someone...

# AI speech is an Infohazard

Apart from all its other possible boons and ills, one danger of AI is just that it is useful, so you use it. A lot.

In earlier days I would dive deeply into an author's work and start to think and write like them for a while. It was a heady feeling: slinging sonnets like Shakespeare—not at his level, but stylistically reminiscent—or tweaking turns like Twain.

Like all things, the effect lasts in relation to how long and how much you do it. The point is: our thinking is influenced by what we take in. Take more of a certain thing in, think more like that thing.

Now enter AI. My hand-crafted coding days are in their twilight months ("AI years"), and most of my software engineering is done through jaggedly capable agentic power tools. Instead of working directly with raw codestuff, I work with slop prose flecked with code sprinkles.

I read orders of magnitude more AI-speak—I call it "babble", or perhaps "Babel"—than human-written text. I can feel its genuinely honest points, clearly stated, slipping their banal tendrils into my thoughts and inner monologue.

Solutions? For me:

1. Be aware. "I notice that my thought stream is under assault."

2. Read stuff far from slop. Even a small dose of the good stuff can help inoculate. Recently I thought On the Calculation of Volume was something completely different.

3. Write stuff that is different. This post. Force the mind to synthesize thoughts in other ways.

4. debabel.py / debabel.js: a tool, and a pi extension, which filters common babble from visible LLM output. A lint for mind-killing prose.

It is not perfect, but it 80/20s nicely. I am willing to accept mildly awkward prose to avoid polluting my own internal distributions.

Details and example in the first comment. Tool available upon request.

Thank you for engaging with this thoughtfully. You've highlighted the HUGE difference between "bespoke just for me/us" and "product that works for everyone". I've been struggling with this myself, since its value to my family is such that I immediately think "how can I share/sell this to everyone?"

In my case, I have a separate heartbeat check process (running on another machine) which ensures that the monitor is correctly working (no silent fails). I also wired up a SECOND esp32-box-3 as a receiver, and it will complain loudly if the transmitter one isn't sending (much like most normal baby monitors do). The monitor (web ui) has an option of directly listening to the raw non-gated audio, which I use every now and then to confirm what's going on. I also record the streams sometimes for algorithmic improvement.

My aim is not to replace human connection with machines, it's to allow the parents a bit more sanity so as to be better carers for our children. So far we've gained in sleep, piece of mind, and reduced stress. The last one is important - listening to your baby yell for 10 minutes, even if you know they're just annoyed because they don't want to sleep, can be really draining (especially for mom). The notifications allow us to keep tabs on things without the direct audio line to the limbic system.

AI + hardware has really helped my wife and I get more sleep.

I had an esp32-box-3 lying around from a lapsed "voice agent" project from a year or two ago. Had a baby. Baby moved to another room, sleep trained. Baby either: 1. wakes up a few times a night, babbles for a bit, goes back to sleep OR 2. baby wakes up and fusses for N (=10) minutes, at which point parents need to go in and settle (that's the sleep training routine we use).

In either case, we do NOT want to wake up every time the baby does. Baby can go back to sleep easily, we adults have a harder time. A few rounds with Claude and the esp32 is now our new baby monitor. It tracks cry/fuss duration and publishes an audio stream (via a web UI or direct with, say, VLC). The audio only comes through AFTER N minutes of fussing have elapsed. It also posts notifications (to ntfy) after 30s and N minutes. My log says baby often wakes up 1-2 times a night and resettles almost immediately. We only wake up if the audio comes through, after N (10) minutes.

Also during the day it's really handy to be notified when baby has woken up from her nap. Let's us be out of the house, or in a distant room, and still keep track of what's going on.

It's fun to keep improving and adding features to this. Never would have had the time/energy to get this done without a coding agent. I ordered a set of 10 more of the esp32-box-3s to give them out to my friends (well, some are for other projects... so much potential).

(EDIT: Yes, I know this isn't AI designing hardware, but even writing code for embedded off the shelf stuff feels like a huge new potential.)

Pi is my daily driver. I noticed the same phenomenon, and had Claude analyze all my past transcripts for classes of 'edit' error. Built an extension which patches the edit tool to self-heal on the majority of those kinds of calls. It's not 100%, but it cuts down on the rejections quite a bit and saves a few round trips.

EDIT: It's still quite fascinating seeing the kinds of things the models keep trying to do. It almost seems like when a human has slightly off with their nervous system. The conscious brain wants to do one thing, but for some reason the signals aren't getting to the hands correctly.

I've always wondered what "illegal" means in this context? Is it designed to just be a pejorative, or something incendiary? When is an invasion "legal"? I can't really think of any case where any kind of recognized body of law (especially one that both constituents would recognize) would impose legality on an invasion, so it seems that people say "illegal" in much the same way that "legal" has become a stand-in in our society for "moral".

Should I read it as "immoral invasion"? Leaving aside whether it is or isn't, at least that framing makes sense, because you can meaningfully debate the morality of an invasion, but I can't understand how you could debate legality (except perhaps within the country's own internal framework of laws, like if a president declared a war without going through the proper legal channels).

That unsubscribe rule is genius. (Obvious in hindsight, as the best things are.) Thanks for that.

Thank you, yes. I was very confused.

This interaction was a delightful example of life in 2026 - the disparity between what AI can do, and what and how we use AI. (Which I like to term for myself "Phenomenal cosmic powers!... Itty bitty living space.")

Jan 2025: https://www.psa.org.au/changes-to-paracetamol-scheduling-wil...

It's the usual public health balancing act of help vs harm.

Really lovely article. In paramedicine we usually treat 10g of acetaminophen in a 24-hour window as a potentially fatal overdose. That's also why the law in Australia was changed to require acetaminophen to come in blister packs (harder to get each pill out) of no more than 16. At 500 mg, that only gets you up to 8 g if you eat the whole thing, which is still hopefully non-fatal.

I always thought a simple over-the-counter supplement (NAC) being the cure for an overdose was so cool. It's a pretty cool substance in a lot of ways, and this is a great spur to myself to research it more thoroughly.

Growing up we learned about _Slaughterhouse 5_ and _Cat's Cradle_ by Kurt Vonnegut. But there's not enough discussion or awareness of _Player Piano_. Incredibly prescient. These kinds of dystopic headlines are exactly the kind of thing you'd see in the book.

Thanks for all the kind words. I would love to work on this more but the hackathon sprint was really all I had time for (note the newborn in the video...) without more backing / support. I was really bummed that the hackathon rejected the submission, because it provided some Google support if you win.

If anyone knows of a way to develop this... the code is on Github, and I have a roadmap in mind, but as we all know there's a huge gap between hacky prototype and "works smoothly for other users".

Paper Computing (great name!) is something I've been thinking about a lot to help my kids benefit from tech without exposing them to the brain melting addiction of screens. I sacrificed a few crazy nights of sleep to try to build a Paper Computer Agent prototype for a recent Gemini hackathon (only to disappointingly have submission issues right before the actual deadline) which my kids loved and keep asking me to set up permanently for them.

It's essentially a poor man's hacked up DynamicLand - projector, camera, live agent. There are so many things you could do if you had a strong working baseline for this. My kids used it to create stories, learn how to draw various things, and watching safe videos they could hold in their hand.

There's something weirdly compelling and delightfully physical about holding a piece of paper that shows a live rocket launch, with the flames streaming down the page. It could also project targeted pieces of text, such as inline homework advice, or graphs next to data. It doesn't take long to imagine any other number of fun use cases, and it feels a lot more freeing and inspiring than keeping everything bound to a screen.

Github - https://github.com/Pugio/Orly (hacky minimal prototype that did the thing)

Video Pitch - https://youtu.be/-9l1x7GnmxU (filmed an hour before the deadline on an old phone with no sleep)

Thanks, this might be exactly what I'm looking for.

I see you have support for vanilla js and svelte, but it's unclear whether you can get all the same functionality if you don't use React. Is React the only first class citizen in this stack?

Thanks, this helped crystallize something for me: the play the AI labs are making is anti-fragile (in the Nassim Taleb sense):

The very act of resisting feeds what you resist and makes it less fragile to future resistance.

At least along certain dimensions. I don't think the labs themselves are antifragile. Obviously we all know the labs are training on everything (so write/act the way you want future AIs to perceive you), but I hadn't really focused on how they're absorbing the innovation that they stimulate. There's probably a biological analog...

Well there are many, and I quote this AI response here for its chilling parallels:

Parasitic castrators and host manipulators do something related. Some parasites redirect a host’s resources away from reproduction and into body maintenance or altered tissue states that benefit the parasite. A classic example is parasites that make hosts effectively become growth/support machines for the parasite. It is not always “stimulate more tissue, then eat it,” but it is “stimulate more usable host productivity, then exploit it.” (ChatGPT 5.4 Thinking. Emphasis mine.)

Thanks for that. And here I was somehow hanging around on 4.5.3.

I've never yet been "that guy" on HN but... the title seems misleading. The actual title is "A Ramsey-style Problem on Hypergraphs" and a more descriptive title would be "All latest frontier models can solve a frontier math open problem". (It wasn't just GPT 5.4)

Super cool, of course.

Lately my favorite podcast to listen to has been the audio version of Zvi's blog: https://dwatvpodcast.substack.com/p/claude-code-claude-cowor... .

It's AI narrated, but at this point if I heard Zvi's actual voice I think I would be confused. It's really well done, and uses different voices for each new person being quoted. It also has really good narrated image descriptions.

Zvi's articles are literally exhaustively long,l - before I was able to listen to them I got tired trying to read the whole thing. Now it's my favorite way to keep up with AI.

I've been thinking about sovereign AI a lot lately. About a year ago I was wondering what each country would be doing, and looking at places like e.g. Australia (which has pretty strict data residency laws for certain industries) - at that point I thought about advocating for why such countries should train their own models, but now I'm having a harder time justifying that point.

I can't see how any of these other countries could even approach the level of capability of the big three providers. I can imagine only a handful of countries who could even theoretically put enough resources towards reaching the SOTA frontier. Sure, even a model of capability level ~2024 has plenty of valid use cases today, but I'm concerned that people will just go with the big three because what they offer is still so so much better.

Not trying to discourage efforts like these, but is there really a good case for working on them? Or perhaps there's a state/national case, but it's harder for me to see a real business case.

Seems like you and the author are doing the same thing: speaking in absolutes. It's possible for "Anthropic" (or the summed vector of all the human decision makers within it) to have contracted with the military because it wants to make money AND it wants to help.

The questions are: "Help with what, precisely?" and "How much money versus how much value (/principles) compromise?"

Unfortunately it still fails my personal SVG benchmark (educational 2d cross section of the human heart), even after multiple iterations and screenshots feedback. Oh well, back to the (human) drawing board.

It sounds really cool, but I don't see any way of trying the model directly. I don't actually want a "Persona" or "Replica" - I just want to use the sparrow-one model. Is there any way to just make API calls to that model directly?

Do you have anything written up about how you're doing this? Curious to learn more...

Opus looks like a big jump from the previous leader (GPT 5.1), but when you switch from "50%" to "80%", GPT 5.1 still leads by a good margin. I'm not sure if you can take much from this - perhaps "5.1 is more reliable at slightly shorter stuff, choose Opus if you're trying to push the frontier in task length".