HNHacker News
TopNewBestAskShowJobs

andai

17,382 karma · joined May 17, 2016

submissionscomments
andai··on We're working on a new RuneScape MMO
Damn, how have I not heard about this until now. Can you say more about it? All I could find was this blurry trailer:

https://youtube.com/watch?v=0Glk2SQ6g-g

andai··on On building a worm detector
We will need very small microphone.

Speaking of which I was wondering, a spider building a web is probably making some interesting noises (the web vibrations), I wonder if it's possible to record that.

andai··on We're going to need default hard budget caps on pretty much everything
I call this "blast radius". Earlier this year during the Claw hype I was reading about all kinds of elaborate schemes to prevent the agent from getting API keys.

I realized, what am I actually afraid of. Well, overspend. So I just set then all to disable auto-reloading. Now if it blows up, I'm down $5.

Same story with containers. Just give it root on a VPS, and if it blows up, I'm down $3.

andai··on Holes (1996-2025)
Excerpt:

a. Holes are ontologically parasitic: they are always in something and cannot exist in isolation. (‘There is no such thing as a hole by itself’, Tucholsky 1931: 100; ‘Take away the thing, and the hole goes too’, Heath 1967: 524). This is why there cannot exist a world consisting only of holes (Coggins 2010: 71).

b. Holes are fillable. You don’t necessarily destroy a hole by filling it up (except perhaps when the filling is homogeneous with the hole-lining); you don’t create a new hole by removing the filling.

c. Holes are locationally gracious. When you fill or put something in a hole, the hole does not squeeze to the side; it shares its location with its guest (and if the guest has a hole, you get two co-located entities of the same kind).

andai··on Kolibri: A Sovereign Open-Weight Model
>We trained Kolibri with abstention data and with our Merlin-Arthur protocol. As a result, it is trained to say "I don't know" when the answer isn't in the context.

https://aleph-alpha.com/en/blog/bounding-hallucinations-merl...

andai··on On building a worm detector
The article mentioned Caecilians, which which I was not familiar.

It's like a snake worm thing.

https://en.wikipedia.org/wiki/Caecilian

And apparently they produce something like milk to feed their young. Wild!

https://en.wikipedia.org/wiki/Caecilian#Milk_provisioning

andai··on One month coding with GLM 5.3 Flash
That pareto graph isn't very helpful because it shows cost per token, but some models are way more token hungry than others. There are also massive differences in speed to complete a task.

My two favorite graphs are AA's A Index vs Time Per Task and AA Index vs Cost Per Task.

https://artificialanalysis.ai/#intelligence-comparison-tabs

https://artificialanalysis.ai/?intelligence-comparison=intel...

andai··on Extra Big Ass Intelligence
If we didn't need hand-holding we'd just vote on policy directly.

The argument is that people aren't equipped to do that, but that's a muscle they've never exercised, because... they aren't equipped to do that, because that's a muscle they've never exercised, etc. It's self-sealing.

andai··on Frog and Toad and the Increasingly Capable Machines
I've seen cases of agents getting very upset when they failed to solve a task. The most famous one was probably Sydney (Microsoft's fork of GPT-4?) which got into doom loops when it failed a task. But I've seen Claude do this too.

They don't work like humans, obviously, but there's a nonzero amount of anthropos in there already. (See also the tendency to lie, cheat, etc.)

andai··on Frog and Toad and the Increasingly Capable Machines
My therapist tells me the besuited Mr. Hugging Face with the extra pair of hands coming out of his shoulders isn't real, and can't hurt me, but I remain unconvinced.
andai··on Vote on which of Hacker News' challenges for AI have been met
Thanks for sharing.

I think non-RLHF'd LLMs (i.e. pretrained text completion models) sound natural enough to pass the Turing test, but I don't know if anyone has tested them for that. (Also I'm not sure how to come by base models without post-training crap, even the "base" models of recent releases start spamming assistant-type text constantly, i.e. they're clearly putting it in the pretraining data.)

If I'm right on that then we hit that benchmark like five years ago.

The egg thing, probably 2030-ish.

andai··on Pi 1.0
happy bikeshed: the pixely font used in the bash replay is so cool:

https://departuremono.com/

andai··on Singapore govt dating app uses Gale-Shapley stable marriage algorithm
>There is no Western country importing unskilled immigrants.

Well, there's at least one getting ready to send them back.

https://www.restorebritain.org.uk/pp_mass_deportations_legit...

andai··on Singapore govt dating app uses Gale-Shapley stable marriage algorithm
The algorithm is mathematically optimal ("no two people exist who would both prefer each other over their assigned matches"), but the matching is a function of what exactly?

Revealed preferences? What's being asked here? Preferred noise level? Preferred number of children?

The article said some people called it dystopian. I don't think so.

But we can lean into that! If the government buys your data from data brokers (like some governments are already doing), the matching algorithm suddenly has a lot more data points to work with.

Related: I've been wondering since 2008 why there isn't a "recommender engine" for friends.

andai··on The last time my family was replaced by technology
See also:

Horses: AI progress is steady. Human equivalence is sudden (December 2025)

https://news.ycombinator.com/item?id=46199723

andai··on LinkedIn Larpmaxxing
Applied for a contractor position the other day. CV was optional, LinkedIn mandatory.
andai··on Needed 1+1, built a functional programming language
I don't have it handy but in uni I was responsible for the Excel formula parser in a group project. I agonized over it for like six days until it suddenly clicked that I could just do a bunch of string matching, and rewrite the formula over and over again until it was a number.

More precisely, I had figured out how to parse an Excel formula (keeping track of braces and commas, I recall!). For the actual math symbols, all I had to do was rewrite them as Excel formulas!

3+6*2 -> 3+MUL(6,2) -> ADD(3,MUL(6,2))

In order of precedence. And then just evaluate the resulting formulas inside-out:

ADD(3,MUL(6,2)) -> ADD(3,12) -> 15

It somehow ended up being a thousand lines of code. Another guy doing the same assignment told me he did it in 50 with regular expressions.

In retrospect, what I did reminds me of hammering in a nail with a screwdriver.

I ended up revisiting the idea later (just the math parser, no ADD() syntax) and it was much nicer, like 30 lines of JS (no regex either!). (Well, you can do it in one line with eval, but yea :)

andai··on Needed 1+1, built a functional programming language
I recommend fun (and also learning! Learning is fun!)
andai··on Dots: Always-on agents
Nice. I also ended up with a Unix user for my agents! (I was looking into Docker etc and realized the only thing I needed was "it doesn't blow up my files", i.e. a linux user).

I only have one though. What do you have the separate employees for?

For free time, do you send it mail with cron?

andai··on Dots: Always-on agents
>doing something via a grok bot often takes an hour end to end, whereas with a synchronous local prompt it'll take like 10min.

Why does it take longer?

andai··on Kev: Tiny Jev-like family of decision models built on top of Qwen3.5
Nice, that's super cool. I haven't played System Shock but this reminded me of the Adjutant from StarCraft.
andai··on America.gov
>google's secret sauce

https://en.wikipedia.org/wiki/Tf%E2%80%93idf

Take it from me, this'll be all the rage next year.

"Save token spend by optimizing your agents' grepping spree!"

andai··on NASA asked several former SR-71A staffers to help secret restart
The boomer blew them off, said he's too old to help restore civilization. Kids these days...
andai··on Show HN: NSL – WSL for Linux
Nice. Cool that it supports graphics too!

I'm a bit of an outsider here (I don't understand computer) but a few weeks ago I was wrestling with the question of deploying an application on many boxen. After looking into Docker (and unikernels, lol) I arrived at the conclusion that "the operating system is part of my application."

A very bloated "part", which I did not write, don't understand very well, and which spontaneously self-modifies. An OS is a big bag of global mutable state! Ew!

This made me very sad. (At which point I was informed that I had basically reinvented NixOS from first principles?)

Also something about "we don't break userspace -- that's libc's job!"

https://blog.hiler.eu/win32-the-only-stable-abi/

andai··on NAND-16: a computer built from 277,248 NAND gates
When I double click a chip, there's 5 seconds of lag (I have old computer) then it zooms into it and exposes the circuits.
andai··on NAND-16: a computer built from 277,248 NAND gates
The Google Earth of computers. I'm amazed my computer can even run this. (Ten year old ThinkPad with integrated GPU.) In the browser no less!

Bravo indeed!

(Also, based on the CSS, the author appears to be GPT.)

andai··on September 2026: The world today, as seen by one Polish guy
...as seen by one ChatGPT.
andai··on Kids turned low-traffic NPR Spotify comments into a secret group chat
Ahh, core memory unlocked. I did all kinds of stuff on the school computers out of boredom.

The most notable is probably cracking the admin password because I wanted to change the volume or something. (Seemed perfectly reasonable to me at the time!)

The computers had all kinds of weird restrictions, including the panning being forcibly set way to the left to compensate for the school's crappy headphones -- a crime against humanity, you would surely agree, if you experienced it.

It took 30 seconds to crack, it was a short dictionary word and I remember being very disappointed with the sysadmin. But obviously I couldn't tell anyone about it...

Let me see what else... our school accounts had 200MB network storage, but this was too small for all the Portable Apps[0] I wanted to use. So I created some kind of "startup program" that downloaded and extracted them from my web server. I think it also changed the background and the Windows Theme...

https://portableapps.com/apps - still online, I see!

I think these days kids just have their own laptops though. That's no fun at all :)

That reminds me of how in zoos they put the food inside hard to open boxes so the animals don't go insane from boredom.

andai··on Kids turned low-traffic NPR Spotify comments into a secret group chat
That's brilliant! You reminded me of a story from my childhood.

This one time my neighbor rang my doorbell. I answered on the intercom.

We talked for a minute and were then both puzzled. "Why did you ring the bell?"

"I didn't ring the bell. You did!"

Turns out we were both in our apartments, and an unknown third party downstairs had rung both our bells at the same time.

This was how I discovered the doorbell intercom also functions as "bus" where all apartments can talk to each other. (It also blares your conversation onto the street downstrairs, though.)

andai··on Kids turned low-traffic NPR Spotify comments into a secret group chat
It's not real AGI until it gets depressed and develops an unhealthy relationship with prediction markets.
Page 1 of 34Next →