fenomas

So ideally that question should have answers that apply to the new question as well.

The point is who decides. If you ask a question and I flag it as a dupe, I might think the answers on the other question apply to yours, but only you know whether they solved your problem or not.

I've no idea if the rate of bad duplicates is so much higher than I observed,

Sure, and neither does SO! They didn't even measure it. They only looked at the signal "does somebody with points think these questions are similar", and discarded the signal of whether the new user got any value out of the site, and I think that's what did them in.

I still think SO was done in by the weird way they handled similar questions. They encouraged veteran users to flag new questions as dupes, even if the "original" question was years old and unanswered. Who does that even help?

Imagine if the system had let veteran users link a new question to an existing answer rather than a question, and if the asker finds it solves their problem they can accept it. At least that way new joiners would have a chance of getting their question answered.

Looking back it feels like SO was one of the first really gamified sites, and the people running it got weirdly focused on the point-economy aspect. They ran the site almost like "points" were a finite resource, and not to be handed out unless the user really deserved it.

I always find it strange how many HN commenters appear to have unconsciously added "exclusively" or "without exception" to the text of the comment they're replying to.

Related, Raymond Chandler says in his letters that he taught himself to write a novelette by copying one (by Erle Stanley Gardner). He took the original story and wrote a detailed synopsis, then wrote a novelette from the synopsis, compared it to the original, did rewrites, and so on until he understood what tricks Gardner had used to make the scenes work.

(fist bump)

I disagree! It's easy to check that an AI program meets its specification, which is to process input tokens and generate output tokens. :)

If you're talking about verifying whether it produces the correct tokens, that's not generally something you can specify in advance with AI. I mean: if your task is one where you can precisely specify which output tokens are correct for a given input, then the task doesn't need AI, no?

I see what you're getting at, but I think you're focusing on the incidental. An animation is good if it's clear and intuitive in motion. It may often be the case that good animations also look nice in a screenshot halfway through, but (a) they don't always, and (b) working well in motion is obviously the more important of the two goals.

As such it makes no sense to worry about the latter thing - it's not a signal whether the animation is good or not.

That's not what TFA is about though.

Look at the youtube example - it has two pieces of UI animating from from a start point to an end point, and the paths are such that they momentarily overlap. There's nothing buggy or janky about it in motion; TFA is just saying that if you ignore the motion and take a screenshot mid-transition it looks odd. Same complaint as what GP describes, and silly for the same reasons.

Nice work, just gave a quick pass but seems to work well!

(Also: vouched, your comment was dead FYI)

Placing an undue emphasis on civility is how bad actors control the conversation.

The load-bearing word in that claim is "undue", and it's not justified here. I'm not doing arcane rules-lawyering, I'm just saying people should avoid doing things the site guidelines quite specifically ask them not to do.

I’m not advocating for this rule to change (I’d appreciate if you didn’t straw man and mischaracterise what I said),

I wasn't suggesting you did, I was suggesting the person I originally replied to might.

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

Does that mean I now repeat your parenthetical back to you? ;)

If you think the guidelines should be changed you can mail dang, but unless they change the civil thing would be to follow them.

Please read what the HN guidelines say about insinuations of astroturfing, because it very much applies here.

(vouched - just fyi this comment was dead, no idea why)

I'm fond of linguistic bugbears, and have actually sent that same article to people before :D But what you're missing is that the less/fewer debate is over their use as adjectives, and TFA's title uses "less" as an adverb. It's asking for AI agents to be less human, not for them to be fewer in number. Swapping it to "fewer" would make the title's meaning no longer match the article.

Now please sit a moment and reflect on what you've done. :P

(The aside was for you - TFA's title is not a case where either word works.)

It's not a grammar issue; only "less" matches TFA's meaning.

(Aside: it's better not to be pedantic, but if you must be pedantic you should remember to be correct as well.)

Nice to have these all collected nicely and sharable. For the amusement of HN let me add one I've become known for at my current work, for saying to juniors who are overly worried about DRY:

Fen's law: copy-paste is free; abstractions are expensive.

edit: I should add, this is aimed at situations like when you need a new function that's very similar to one you already have, and juniors often assume it's bad to copy-paste so they add a parameter to the existing function so it abstracts both cases. And my point is: wait, consider the cost of the abstraction, are the two use cases likely to diverge later, do they have the same business owner, etc.

Nope, "less" is what TFA means.

Curmudgeonly comment from someone trying to sound like a wise elder about how actually all this was the norm even in the days of Usenet.

Not at all - didn't mean to sound snarky, I just wanted to add that I was omitting details and caveats.

FWIW, personally I think it muddies things to frame the question as if "..using statistical token generation" was a limitation. NNs are Turing-complete, so what LLMs do can just be considered "computation" - the fact that they compute via statistical token generation is an implementation detail.

And if you're like most people, "can cognition happen via computation?" is a less controversial question, which then puts LLMs/cognition topics easily into the "in principle, obviously, but we can debate whether it's achievable or how to measure it" category.

The high-order bit for for each case is the category it's in and the "Outcome" column - that summarizes if the solution was full/partial/wrong, if AI had assistance, etc. Then further discussion for each one is linked from the number.

Then the "Literature result" columns have a citations for where similar published results were found. The ones with no "Literature" column, like in the first section, are cases where no similar published results have been found (implying that the solution would not have been trained on). Note that in some cases a published solution was found but it wasn't similar to the AI's.

(this is all explained with more detail and caveats at the top of the page)

What you're asking for is exactly what's in the link you replied about. It collects analysis of each solution (or attempt), and info about whether the AI's solution could be found anywhere in the literature.

generous interpretation

The term of art for that is steelmanning, and HN tries to foster a culture of it. Please check the guidelines link in the footer and ctrl+f "strongest".

The post you replied to was:

We went from 2 + 7 = 11 to "solved a frontier math problem" in 3 years, yet people don't think this will improve?

All that says is that the speaker thinks models will improve past where they are today. Not that it's a logical certainty (the first thing you jumped on them for), and certainly not anything about "limitless potential for growth" (which nobody even mentioned). With replies like this, invoking fallacies and attacking claims nobody made, you're adding a lot of heat and very little light here (and a few other threads on the page).

It's not a side effect of tokenization per se, but of the tokenizers people use in actual practice. If somebody really wanted an LLM that can flawlessly count letters in words, they could train one with a naive tokenizer (like just ascii characters). But the resulting model would be very bad (for its size) at language or reasoning tasks.

Basically it's an engineering tradeoff. There is more demand for LLMs that can solve open math problems, but can't count the Rs in strawberry, than there is for models that can count letters but are bad at everything else.

LLMs are bad at arithmetic and counting by design. It's an intentional tradeoff that makes them better at language and reasoning tasks.

If anybody really wanted a model that could multiply and count letters in words, they could just train one with a tokenizer and training data suited to those tasks. And the model would then be able to count letters, but it would be bad at things like translation and programming - the stuff people actually use LLMs for. So, people train with a tokenizer and training data suited to those tasks, hence LLMs are good at language and bad at arithmetic,

This is like saying chess engines don't actually "play" chess, even though they trounce grandmasters. It's a meaningless distinction, about words (think, reason, ..) that have no firm definitions.

Those are emoticons ;)

Emoji originally came from Docomo phones in Japan around 1999. (Or I think those were the first ones actually called "emoji"; some other earlier devices had similar character sets.)

They would have if they could have - answered here:

https://news.ycombinator.com/item?id=47256093

I remember using that tool internally! Personally I think I only used it to get stats of which features/APIs were popular. But I think other teams used it for QA/conformance, like finding content that occurred in the wild but wasn't covered by test cases.