BalinKing

6 ms·

31.
by BalinKing 5mo ago
From the site guidelines ( https://news.ycombinator.com/newsguidelines.html ): > Be kind. Don't be snarky. Converse curiously; don't cross-examine. Edit out swipes.
32.
by BalinKing 5mo ago
I'm not the other commenter (and I believe you that it's not AI), but I'd guess it's mostly the first line: a short affirmation followed by "The problem is ...." feels like the sort of formula the LLMs love to
33.
by BalinKing 5mo ago
In general, I think this phenomenon is called "phono-semantic matching": https://en.wikipedia.org/wiki/Phono-semantic_matching .
34.
by BalinKing 5mo ago
It's just that, in my (uninformed) opinion, Anthropic is incentivized a priori to claim things like this about their models. Like, it's probably really good marketing to say "our product is so smart, and we're so concern
35.
ImpossibleBench: Measuring LLMs' Propensity of Exploiting Test Cases(arxiv.org)
2 points by BalinKing 6mo ago|0 comments
36.
by BalinKing 6mo ago
In the paper itself, the abstract actually does have a paragraph break, so it's probably just an autoformatting issue or something.
37.
by BalinKing 7mo ago
AI-generated comments are disallowed by HN guidelines: https://news.ycombinator.com/newsguidelines.html#generated
38.
by BalinKing 7mo ago
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
39.
by BalinKing 7mo ago
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 f
40.
by BalinKing 7mo ago
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)
41.
by BalinKing 7mo ago
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)…
42.
by BalinKing 7mo ago
Fun name: seems like a reference to “Magit” both syntactically (being a portmanteau of “Magit” and “jujutsu”) and semantically ( majutsu meaning “magic” in Japanese).
43.
by BalinKing 7mo ago
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 .
44.
by BalinKing 7mo ago
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.
45.
by BalinKing 7mo ago
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？
46.
by BalinKing 7mo ago
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.
47.
by BalinKing 7mo ago
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 l
48.
by BalinKing 7mo ago
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
49.
by BalinKing 8mo ago
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.
50.
by BalinKing 8mo ago
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
51.
by BalinKing 8mo ago
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 resu
52.
by BalinKing 8mo ago
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 mac
53.
by BalinKing 9mo ago
Yeah, I'm on macOS (although even back on Windows, I used to use the Character Map all the time).
54.
by BalinKing 9mo ago
> 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
55.
by BalinKing 9mo ago
See my other comment—LangRust.lean is the same way. EDIT: Just skimmed Completeness.lean, and it looks similar—at a glance, even the 3+-line proofs are very short and look a lot like boilerplate.
56.
by BalinKing 9mo ago
I also took a look at the `LangRust.lean`, and the majority of the proofs are just `rfl` (after an `intros`)—that's a major red flag, since it means the "theorems", like those in SSOT.lean, are true just by unfolding defini
57.
by BalinKing 9mo ago
The file SSOT.lean is completely trivial, I think: Unfolding the definitions in the theorems, they say nothing but "x=1 => x=1", "x=1 => x≤1", and "x≠1 => x=0 ∨ x>1" (where x is a natural number). B
58.
by BalinKing 9mo ago
Keyboards are highly deterministic. And when they're not, e.g. due to physical wear or software glitches, this makes them basically unusable for touch typists.
59.
by BalinKing 9mo ago
Off the top of my head, I want to say you can right-click on the current folder name to see (and navigate to) all its ancestors.
60.
by BalinKing 9mo ago
I'm fairly skeptical of tests that are closed-book. IMO the only reasons to do so are if 1) the goal is to test rote memorization (which is admittedly sometimes valuable, especially depending on the field) or, perhaps more commonly, 2)

More ›