HNHacker News
TopNewBestAskShowJobs

sundarurfriend

5,212 karma · joined November 26, 2008

A student of life. A lover of programming. A seeker of knowledge. A believer in goodness. A hacker at heart.

@digital_carver@qoto.org

meet.hn/city/11.0018115,76.9628425/Coimbatore

Socials: - https://x.com/digital_carver - bsky.app/profile/sundarar.bsky.social

Interests: Scientific Computing, AI/ML, Books, Education, Freelancing, Open Source, Science, UI/UX Design, Writing, Yoga

---

submissionscomments
sundarurfriend··on How Lewis Carroll computed determinants (2023)
The original comment was about one language that borrowed cipher from Arabic (i.e. English) where the word no longer means zero. So my comment was about a different language that also borrowed the word cipher (i.e. Tamil) where it still retains that meaning.
sundarurfriend··on How Lewis Carroll computed determinants (2023)
In Tamil, it still means a zero. It's usually pronounced like 'cyber' though, because Tamil doesn't have the 'f'/'ph' sound natively.
sundarurfriend··on Mattermost restricted access to old messages after 10000 limit is reached
In what sense?
sundarurfriend··on Show HN: HN Wrapped 2025 - an LLM reviews your year on HN
Pretty fun overall. The roasts were really good, the stats were pretty neat, the predictions made me smile. The xkcdupe fell flat for me because it's pretty generic and I've seen many variants of it before, but still pretty impressive for an AI to make. The "Your HN Front Page in 2035" was definitely the weakest part for me. Maybe because every gag in it had already been done in other sections.

One thing I noticed is that it seems to give more weight to submissions (i.e. things I've submitted as posts) than to comments (or at least, doesn't let submissions get drowned in a sea of comments), which turns out as a good thing. And it doesn't seem to care about karma, which is also IMO a really good thing, makes it feel like a deep dive and keeps it interesting.

sundarurfriend··on T5Gemma 2: The next generation of encoder-decoder models
This made me compare the figures, and: did they accidentally switch those around, or are the Post-training Reasoning and Factuality scores actually significantly lower than the Pre-training ones?

Edit: Just noticed

> Also note pre-training and post-training benchmarks are different, so scores are not comparable across plots.

The paper gives more details about the specific benchmarks and the scores obtained in them: https://arxiv.org/html/2512.14856v1#S4

sundarurfriend··on Beginning January 2026, all ACM publications will be made open access
So that's what this wording means:

> Making the first 50 years of its publications and related content freely available expresses ACM’s commitment to open access publication and represents another milestone in our transition to full open access within the next five years.

( from https://www.acm.org/articles/bulletins/2022/april/50-years-b... )

I wouldn't have understood that nuance without the context given by your comment, but in my developer mind I analogize "freely available" to a "source available" license that they took on, as a step towards going open access ("free and open source") over time. I'm also happy to see that that transition seems on track as planned.

sundarurfriend··on Beginning January 2026, all ACM publications will be made open access
> A lot of these are the underpinnings of modern computer science optimizations.

Note that older articles have already been open access for a while now:

> April 7, 2022

> ACM has opened the articles published during the first 50 years of its publishing program. These articles, published between 1951 and the end of 2000, are now open and freely available to view and download via the ACM Digital Library.

- https://www.acm.org/articles/bulletins/2022/april/50-years-b...

sundarurfriend··on Slowness is a virtue
> The problem is that research work generally takes the following form:

> * Come up with 5 possible approaches (2 days)

Even that gives you something to talk about, that looks "solution-oriented" in the way managers like; in deep work the first steps can actually be:

* Break down the existing approach into its fundamental assumptions and components (3 days)

* Try to build a few approaches challenging some assumptions in a useful way, and fail (2 days)

* Get a clearer picture of the cloud of possibilities that are more likely to work, and start assembling the pieces from that cloud (3 days)

and so on. This is the kind of stuff that's really hard to communicate, and often sounds like you're doing nothing at all in the initial stages, even though that time pays for itself many times more once(/if) you get to take it to completion instead of doing patchwork upon patchwork on a design based on outdated assumptions.

sundarurfriend··on In Defense of Matlab Code
As your comment already hints at, using Python often ends up a hodgepodge of libraries and tools glued together, that work for their limited scope but show their shaky foundations any time your work is outside of those parts. Having worked with researchers and engineers for years on their codebases, there is already too much "throw shit at the wall and see what sticks" temptation in this type of code (because they'd much rather be working on their research than on the code), and the Python way of doing things actively encourages that. Julia's type hierarchies, integrated easy package management, and many elements of its design make writing better code easier and even the smoother path.

> I don't think Julia really solves any problems that aren't already solved by Python.

I don't really need proper furniture, the cardboard boxes and books setup I had previously "solved" the same problems, but I feel less worried about random parts of it suddenly buckling, and it is much more ergonomic in practice too.

sundarurfriend··on GNU recutils: Plain text database
Has support for recfiles improved recently? I was overjoyed when I learnt that VisiData had support for recfiles since I'd been getting tired of editing recfiles manually and liked VisiData's UI when working with CSVs and such.

But at least at the time (~2 years ago IIRC), the support was really basic, just basic record display, and most importantly, editing the parts VD didn't understand lead to data loss. I don't remember what I was trying to do - the error report I wrote with those details died with my old machine - but it wasn't anything too complicated, just array fields and foreign keys I believe i.e. just using recfile features one step beyond a listing of `key: value` pairs. I gave up on recfiles as a whole after losing data a few times like this (since I hadn't found any other suitable tool either).

sundarurfriend··on Hashcards: A plain-text spaced repetition system
I wish there were more/better tools for working with recutils. I had a phase of trying to use recutils wherever it made sense, a few years ago, but the format has a lot of redundancy (not a bad thing in itself), and editor support to make working with that easier was basically non-existent (perhaps it exists only for Emacs). Using the command-line interface for everything was way too cumbersome. Visidata claimed to support the format, which got me excited, but in my experience it mangled the file if you had anything more than a basic set of records, and the support for display too was overall very rudimentary.
sundarurfriend··on GNU Unifont
And just in case the wording "special forms" makes someone think these are edge cases or just fancy-but-non-essential ligature type stuff, these are basic core syllables and sounds in these languages that go from their familiar form to a form where you have to mentally do some detective work to figure out what's being said. 1t'5 1!k3 !f 3v3ry 0th3r w0r<| w@5 wr!tt3n l!k3 th!5. (Not a great analogy but it's the best I got.)
sundarurfriend··on Gut microbial imbalance can impact memory, says study
Seems most likely this one: "Gut dysbiosis leads to cognitive decline through CNTF-mediated activation of microglia in mice" https://link.springer.com/article/10.1186/s12915-025-02454-x

The underlying mechanism seems applicable to humans too, but it's annoyingly ironic that the paper mentions mice in the title clearly and doesn't even mention humans, and the article doesn't mention mice even once.

sundarurfriend··on GPT-5.2
ChatGPT web UI was also like this for the longest time, until a few months ago: all sorts of random UI bugs leading either to data loss or misleading UI state. Interrupting still is very flaky there too. And on the mobile app, if you move away from the app while it's taking time to think, its state would somehow desync from the actual backend thinking state, and get stuck randomly; sometimes restarting the app fixes it, sometimes that chat is that unusable from that point on.

And the UI lack of polish shows up freshly every time a new feature lands too - the "branch in new chat" feature is really finicky still, getting stuck in an unusable state if you twitch your eyebrows at wrong moment.

sundarurfriend··on GPT-5.2
That's interesting, I just today started getting some "Some sites restrict our ability to check links." dialogue in ChatGPT that wanted me to verify that I really wanted to follow the link, with a Learn More link to this page: https://help.openai.com/en/articles/10984597-chatgpt-generat...

So it seems like ChatGPT does this automatically and internally, instead of using an indirect check like this.

sundarurfriend··on GPT-5.2
> new context management using compaction.

Nice! This was one of the more "manual" LLM management things to remember to regularly do, if I wanted to avoid it losing important context over long conversations. If this works well, this would be a significant step up in usability for me.

sundarurfriend··on GPT-5.2
Are you saying ChatGPT's voice chat is of good quality? Because for me it's one of its most frustrating weaknesses. I vastly prefer voice input to typing, and would love it if the voice chat mode actually worked well.

But apart from the voices being pretty meh, it's also really bad at detecting and filtering out noise, taking vehicle sounds as breaks to start talking in (even if I'm talking much louder at the same time) or as some random YouTube subtitles (car motor = "Thanks for watching, subscribe!").

The speech-to-text is really unreliable (the single-chat Dictate feature gets about 98% of my words correct, this Voice mode is closer to 75%), and they clearly use an inferior model for the AI backend for this too: with the same question asked in this back-and-forth Voice mode and a normal text chat, the answer quality difference is quite stark: the Voice mode answer is most often close to useless. It seems like they've overoptimized it for speed at the cost of quality, to the extent that it feels like it's a year behind in answer reliability and usefulness.

To your question about competitors, I've recently noticed that Grok seems to be much better at both the speech-to-text part and the noise handling, and the voices are less uncanny-valley sounding too. I'd say they also don't have that stark a difference between text answers and voice mode answers, and that would be true but unfortunately mainly because its text answers are also not great with hallucinations or following instructions.

So Grok has the voice part figured out, ChatGPT has the backend AI reliability figured out, but neither provide a real usable voice mode right now.

sundarurfriend··on A “frozen” dictionary for Python
Can someone ELI5 the core difference between this and named tuples, for someone who is not deep into Python? ChatGPT's answer boiled down to: unordered (this) vs ordered (NTs), "arbitrary keys, decided at runtime" vs "fixed set of fields decided at definition time" (can't an NT's keys also be interpolated from runtime values?), and a different API (`.keys()`, `.items()`), etc (I'm just giving this as context btw, no idea if there's inaccuracies in these).

So could this also have been approached from the other side, as in making unordered NamedTuples with support for the Mapping API? The line between dictionaries and named tuples and structs (across various languages) has always seemed a bit blurry to me, so I'm trying to get a better picture of it all through this.

sundarurfriend··on Icons in Menus Everywhere – Send Help
> "ancient disk icon"

Even though floppy disks were a bit before my time and I rarely ever used them, seeing them be called ancient disk makes me wanna find the nearest coffin and just go lie down. :D

sundarurfriend··on Tiny Core Linux: a 23 MB Linux distro with graphical desktop
> puppy Linux, which I could never get into

In what way? Do you mean you didn't get the chance to use it much, or something about it you couldn't abide?

sundarurfriend··on HTML as an Accessible Format for Papers (2023)
Do the older papers work via [Ar5iv](https://ar5iv.labs.arxiv.org/) ?

> View any arXiv article URL [in HTML] by changing the X to a 5

The line

> Sources upto the end of November 2025.

sounds to me like this is indeed intended for older articles.

sundarurfriend··on HTML as an Accessible Format for Papers (2023)
[Sept 2023] as per the wayback machine.
sundarurfriend··on Autism should not be treated as a single condition
> Instead, we ask whether drinking often interferes with taking care of home and family; or leads to job/school troubles; or has lead to getting arrested.

We don't ask just that, and the diagnosis doesn't hinge on those - in fact those account for only 3 (or 4 depending on how you count) of the 11 diagnostic criteria for alcohol use disorder. The others are about the person's own experience with alcohol, the difficulties and psychological problems caused by it to the person themself. And that's for alcohol use, an external behaviour-based problem with a specific narrow scope. Autism is a much wider construct with much more varied impact and experiences, and yet in practice people are placed somewhere on the spectrum based mainly on external interactions and troubles.

Historically this came about because people who were "low-functioning" caused more difficulties to others, whereas "high-functioning" folk didn't - even though they might have comparable amounts of difficulties and psychological anguish internally and in need of similar help too. This simplistic view is changing slowly within the field and with some therapists recognizing it better for what it is, but it's still not nearly as widely recognized as it needs to be.

sundarurfriend··on Autism should not be treated as a single condition
I suspect part of your parent comment's point is that this is an implicit bias in the way the spectrum is defined and thought of, so it wouldn't be clinically defined in those terms explicitly.

In other words, the "spectrum" doesn't exist to capture the variation in the autistic person's own experience - if it did, it would look very different. It's a remnant of a time when autism was seen as just a "problem" for the people around you, and the spectrum measures how much of a problem you are and how weird you are seen by their measure; which does map onto a continuous line in the same way.

That does capture something useful, but only a small part of what autism actually comprises, and is much less useful at capturing the autistic person's own experience of it, and makes it a less useful tool to them than people might assume.

sundarurfriend··on Ghostty is now non-profit
Yes, as I mentioned, Slack made a U-turn after the thread got popular and became a PR disaster (after having ignored the issue again and again previously). That kind of behaviour indicates a service that shouldn't be relied on so much, at least not without low-friction alternates that you keep ready to jump onto, in case some exec there decides again that the Club is too tempting a prize to not attempt another squeeze.
sundarurfriend··on Ghostty is now non-profit
> The students have one big global Slack instance.

Are you back on Slack as the primary comms channel after their sudden attempt to upcharge you (followed by the U-turn after the PR backlash)? Do you have some mirroring and other kind of fallback strategy if something like that happens again?

(Context for those who missed it: https://news.ycombinator.com/item?id=45283887)

sundarurfriend··on Zig quits GitHub, says Microsoft's AI obsession has ruined the service
You can ignore them yourself, but people will get annoyed, take to social media, post "this project ignores PRs" "dead project" "they're not inclusive" etc. without context, and those things can gain momentum out of nowhere and have real consequences. And there will goalpost-shifting again, from "just mirror to GH" to "just accept PRs on GH too, why are they being so obstinate" and then further along.

The Linux kernel is unlikely to get those and can survive that even if it does, but your average FOSS project can be severely damaged in terms of reputation, contribution, usage, etc., by this kind of stuff, even if you want to dismiss it as just "social media fluttering". Easier and safer just to keep things clean and clearly at one place.

sundarurfriend··on Zig quits GitHub, says Microsoft's AI obsession has ruined the service
One problem is that GH gives you no way to disable PRs. And even if you write in BIG BOLD LETTERS that PRs should be on Codeberg not GH, people get upset and make a fuss over their "ignored" PRs and it ends up making unnecessary headaches for you over and over.
sundarurfriend··on Zig quits GitHub, says Microsoft's AI obsession has ruined the service
I hope Codeberg is able to scale up to this surge in interest, but

> it is just a single binary that takes a minute to setup - and setup a local mirror of your Codeberg repo with code, issues, etc so you have access to your issues, wiki, etc

is really cool! Having a local mirror also presumably gives you the means to build tools on top, to group and navigate and view them as best works for you, which could make that side of the process so much easier.

> you'll have to update them manually later

What does the manually part mean here? Just that you'll have to remember to do a `forgejo fetch` (or whatever equivalent) to sync it up?

sundarurfriend··on Nvidia announces new open AI models and tools for autonomous driving research
Related paper published a month ago: https://arxiv.org/abs/2511.00088
← PreviousPage 2 of 34Next →