evertedsphere

Another day, another LLM-generated blog post on the front page about a deeply human topic. Do others not detect the tells, or do you not care?

(And, no, there is no "respond to content rather than style" issue here. There is no meaningful content here. That would be the prompt, but of course the author doesn't want to just post that.)

that's pretty good! i'll take a look at converting some small add-ons using it later

the docs were clearly not written by a human, so proceeding from the expectation of global coherence, or that of adherence to principles that would be implied to a human being reading this in a certain context, is likely to end up being a waste of time

i don't think this can be called a "jailbreak"

it's a prompting "style" that works over a long exchange

excellent idea. few things feel like they need to be extensions, and this would also make it easy to hack on stuff like the zotero connector and maintain control over random tiny unmaintained extensions.

first things that come to mind: can every extension be converted? what do i lose in terms of usability, if anything? in a simpler time, i would have expected these things to be at the top of the docs if there were docs, or else for the docs to just be a nearly empty readme containing just "# TODO\nwrite readme"

repeat after me: if your readme wasn't worth writing, it isn't worth reading

the readme is the public face of the software that you are posting to this site for the consideration of potential users. why start by disrespecting them with a list of "Key Features" (linking to an equally doubtful-looking architecture document) that's just raw data that you didn't see fit to arrange into an introduction based on your understanding of what a human user might want?

even assuming it's all correct, which it might be if you've proofread it, nowhere in this impersonal wall of token vomit is there consideration for an actual user, because the purpose of these tools is management-brained report generation—and that is not what documentation is

in case someone from anthropic is reading this: could you please add a bit of padding on the outside of the page? at least on a phone screen, the text covers the entire width of the screen from edge to edge

what is that bar (= token span) on the right common to the first three models

please. no more chatgpt-generated readmes upvoted to the front page

if you couldn't be bothered to write it, why should anyone read it, and what does that say about your view of the potential users you're trying to attract

also, you can leave a comment on your own post highlighting what you found interesting about the link

rust-style "sum type" errors are values too

until the day when the person/project/company whose code it breaks has a sufficient amount of pull over you/your team that it becomes your problem

that's why you prevent it from ever coming into existence if you can

yes, it's the equivalent of |AB| for a line segment and is not uncommon in high school maths texts in some parts of the world

> What about limiting the number of tickets per card?

discussed in TFA

> Or do what airlines do and you need to declare who is using the ticket

ditto

ruler https://arxiv.org/abs/2404.06654

nolima https://arxiv.org/abs/2502.05167

the rest of the title of the linked post is in latin

would appreciate if you could post the prompt as well so the rest of us could learn how to generate our hn comments too

that's what i said

https://en.wiktionary.org/wiki/audi#Latin

or, if you prefer, https://en.wiktionary.org/wiki/Audi#German

the title reminded me that suno is hindi for audi

the evidence keeps pouring in for why it's a good idea to be running llms locally if you're going to put a business on top of one (or have enough money to have a contract where the provider can't just do this, but i don't know if that is a thing unless you're running on-prem)

you could also remove those pieces from the react examples

it's hn greying out the post

an undergraduate degree in a field is not enough to understand recent research in a specialised subfield of a subfield and you shouldn't beat yourself up over that

there's nothing wrong with you, you just need the right background and you can go get that. see e.g. the fast.ai course

can we just ban llm-generated blogspam submissions already

why not just post the prompt you used and let readers feed it into the language model themselves

as seen also in Spätzle, Müsli, or, to pick something more relevant on HN, the words Brötli (or Zöpfli)

-li is a different version of the same ending

"Show HN: Automatic HN Prediction Market"

0-10 inclusive is one extra even

how do you tolerate the sheer latency of running the "vast majority" of your web searches through an llm