Hacker News: BalinKing

ImpossibleBench: Measuring LLMs' Propensity of Exploiting Test Cases

BalinKing — Wed, 25 Mar 2026 19:20:28 +0000

Article URL: https://arxiv.org/abs/2510.20270

Comments URL: https://news.ycombinator.com/item?id=47521953

Points: 2

# Comments: 0

New comment by BalinKing in "Meta's Omnilingual MT for 1,600 Languages"

BalinKing — Sat, 21 Mar 2026 18:35:10 +0000

In the paper itself, the abstract actually does have a paragraph break, so it's probably just an autoformatting issue or something.

New comment by BalinKing in "Avoiding Trigonometry (2013)"

BalinKing — Thu, 12 Mar 2026 14:58:45 +0000

AI-generated comments are disallowed by HN guidelines: https://news.ycombinator.com/newsguidelines.html#generated

New comment by BalinKing in "Ask HN: Please restrict new accounts from posting"

BalinKing — Sun, 08 Mar 2026 20:19:02 +0000

I think a steelman interpretation of the parent is that entirely LLM-generated projects should be disallowed. There's a lot of submissions on Show HN that seem completely vibe-coded to me (like, including the README), which is a very different situation IMO from someone who simply used Claude to write some—or even most—of the code. When even the human-facing portion of a submission is LLM-generated, it bothers a lot of people (myself included).

New comment by BalinKing in "Ask HN: Please restrict new accounts from posting"

BalinKing — Sun, 08 Mar 2026 20:15:01 +0000

Sorry, updated my original comment—I meant to qualify it to only those cases where it's blatantly obvious. Obviously a lot of ambiguous comments will slip through as a result, but I agree with you that false negatives are better than false positives.

New comment by BalinKing in "Ask HN: Please restrict new accounts from posting"

BalinKing — Sun, 08 Mar 2026 19:41:41 +0000

I furthermore wish that "posting an LLM-generated comment (i.e. and passing it off as your own)" was worthy of an instant ban, because I see this sort of behavior from non-green accounts as well.

EDIT: I meant (but totally forgot) to qualify that my "proposal" would only apply when the LLM-ness is self-obvious—idk, make up a "reasonable person" standard or something. Presumably, the moderators would err on the side of letting things slide. Even so, many comments I've seen are simply impossible for any reasonable person to claim as "human-written"—the default ChatGPT style is simply too distinct.

New comment by BalinKing in "Bourdieu's theory of taste: a grumbling abrégé (2023)"

BalinKing — Sat, 07 Mar 2026 17:49:04 +0000

Slightly surprised to learn Master and Commander is “lowbrow”—is it just because it’s not an art film or whatever? Usually I’d expect Marvel films to be described that way (unfairly imo, when it comes to the Phase One batch at least)…

New comment by BalinKing in "Editing changes in patch format with Jujutsu"

BalinKing — Sat, 07 Mar 2026 17:38:38 +0000

Fun name: seems like a reference to “Magit” both syntactically (being a portmanteau of “Magit” and “jujutsu”) and semantically (majutsu meaning “magic” in Japanese).

New comment by BalinKing in "Working and Communicating with Japanese Engineers"

BalinKing — Sat, 07 Mar 2026 15:48:10 +0000

Minor nitpick, but I didn’t think テーマ (tēma, “theme”) was an abbreviation—Jisho and Wiktionary (for what they’re worth) say it’s from German Thema.

New comment by BalinKing in "Don't make me talk to your chatbot"

BalinKing — Wed, 04 Mar 2026 00:57:44 +0000

In that case, wouldn't you be happy to get more calls, so that the up-front "training" cost is worth it? Naïvely I'd expect that every additional call would _decrease_ the amortized price per call.

New comment by BalinKing in "Show HN: Tomoshibi – A writing app where your words fade by firelight"

BalinKing — Sat, 28 Feb 2026 19:54:56 +0000

ちなみに、日本語のバージョンでイタリック体の漢字と仮名があることに気づきましたが、それが普段ですか？僕は日本語が下手ですが、日本語でイタリック体の字があまり使われていないと聞いたことがあるだけです。でも、やはり実践にそうじゃないですか？

New comment by BalinKing in "The peculiar case of Japanese web design (2022)"

BalinKing — Tue, 24 Feb 2026 00:18:22 +0000

To be fair, Japanese headlines use a specific writing style that is much more compressed than normal text, like how English newspaper headlines drop words like “a” or “is” to save space.

New comment by BalinKing in "15 years later, Microsoft morged my diagram"

BalinKing — Wed, 18 Feb 2026 14:43:52 +0000

The apples one is LLM nonsense: the left example doesn’t include any code for the loop, whereas the streams version actually is iterating over a collection.

Regardless, FP-style code isn’t “shiny new stuff”—it’s been around for decades in languages like Lisp or Haskell. Functional programming is just as theoretically “fundamental” as imperative programming. (Not to mention that, these days, not even C corresponds that closely to what’s actually going on in hardware.)

New comment by BalinKing in "Claude Sonnet 4.6"

BalinKing — Tue, 17 Feb 2026 21:59:39 +0000

Sonnet 4.6 gives me the fairly bizarre:

> Walk! It would be a bit counterproductive to drive a dirty car 50 meters just to get it washed — and at that distance, walking takes maybe 30–45 seconds. You can simply pull the car out, walk it over (or push it if it's that close), or drive it the short distance once you're ready to wash it. Either way, no need to "drive to the car wash" in the traditional sense.

I struggle to imagine how one "walks" a car as distinct from pushing it....

EDIT: I tried it a second time, still a nonsense response. I then asked it to double-check its response, and it realized the mistake.

New comment by BalinKing in "Mathematicians disagree on the essential structure of the complex numbers (2024)"

BalinKing — Tue, 10 Feb 2026 19:50:45 +0000

I'm presuming this is old news to you, but what helped me get comfortable with ℂ was learning that it's just the algebraic closure of ℝ.

New comment by BalinKing in "Is particle physics dead, dying, or just hard?"

BalinKing — Tue, 10 Feb 2026 04:51:51 +0000

There are various theories about what's actually happening in quantum mechanics. Some theories have hidden variables, in which case the issue is simply one of measurement (i.e. there really is an "objectively correct" value, but it only looks to us like there isn't).[0] However, this is not known to be the case, and many theories really do claim that position and momentum fundamentally cannot both be well-defined at once. (The "default" Copenhagen interpretation is in the latter camp; AFAIK it's convenient in practice, and as a result it's implicitly assumed in introductory QM classes.)

[0] Well, and the hidden variables are non-local, which is a whole 'nother can of highly non-intuitive worms.

New comment by BalinKing in "Typechecking is undecidable when 'type' is a type (1989) [pdf]"

BalinKing — Mon, 02 Feb 2026 00:44:54 +0000

My 2¢ from an interactive theorem proving perspective: In so-called computational type theory, typechecking is indeed allowed to be undecidable, and you get a lot of cool expressive power (e.g. well-behaved quotients and subtypes) as a result. This was one of the big ideas behind NuPRL back in the day, and Istari[0] more recently.

[0] www.istarilogic.org

New comment by BalinKing in "The longest Greek word"

BalinKing — Sun, 18 Jan 2026 15:08:46 +0000

From what I've read, the German phenomenon isn't actually German-specific after all, and English does it too; the difference is just that English keeps the spaces when written. Like, linguists apparently consider "vending machine" to be a perfectly cromulent compound word (among other things, consider that the stress falls on "vending" instead of "machine," which wouldn't(?) happen if "vending" was being used as a bona fide standalone word). Turns out, there's not even an accepted general definition of what a "word" even is in the first place, because different languages vary so much.

A slightly more thorough discussion from an actual linguist: https://youtu.be/tfnANe2YUwM?si=LAxriH-RuqmUgrxl.

New comment by BalinKing in "Show HN: Lean4 proof that SSOT requires definition-time hooks and introspection"

BalinKing — Thu, 08 Jan 2026 17:32:33 +0000

Yeah, I'm on macOS (although even back on Windows, I used to use the Character Map all the time).

New comment by BalinKing in "Show HN: Lean4 proof that SSOT requires definition-time hooks and introspection"

BalinKing — Thu, 08 Jan 2026 17:31:10 +0000

> Compare to software verification: when you prove a sorting algorithm correct, the hard work is the loop invariants and the model, not the final QED. Tedious proof steps usually indicate you're fighting your abstractions.

This is a false statement when working with an interactive theorem prover like Lean. Even trivial things require mountains of effort, and even blatantly obvious facts will at least require a case analysis or something. It's a massive usability barrier (and one that AI can hopefully help with).