Philosophical question regarding blacklisted websites. I'm looking at one with ~660 TP, 9 FP, and 1 NAA feedback. With some negative look-behinds, I can get that down to 0 FP, and maybe clear or ask for the one NAA feedbacker to revisit. Would we rather do that (ugly regex on blacklist) or downgrade it to a watch-list?

@JeffSchaller If it's reasonable to think we'll see a detection at some point, then I really don't care how ugly the regex is. If it isn't all that high-value, then... Eh, downgrade it.

Considering we're looking at around 660 TP, then I'd say it's probably worth keeping in some form.

it's a messy case of a TLD having lots of spam, but the occasional innocent post asking something like "how do I block $X" or "parse this json data that happens to have $X".

I'm wondering if the rule would be: "ugly regexes are OK, as long as you don't lose any TP", but I don't want to make the rules :)

Look there are some really unsightly regexes on the lists -- in my opinion, anyway - I'm sure Makyen would look at some of the truly unholy ones and go "That's perfectly readable idk what you're talking about" ;)

If it works, it works. :D

true! ugly is in the eye of the beholder regex author :)

yeah I see no issue with ugly regexes

for the example here, I just kept adding negative look-behinds for the FP; maybe that's OK?

660/9/1 is still pretty blacklist-worthy, honestly

oh, yeah, that's not even bad; a list of negative look-behinds is a pretty standard trick

see, e.g.,

it actually is (blacklist-worthy), because I picked the wrong place to start in my spreadsheet; I went to the bottom (near the 99% TP) instead of to the < 95% TP's. But, the question stands as far as how much effort/regex to put into keeping a blacklist entry over downgrading it

I mean, technically it should have 100% TP to be on the blacklist, but I'm good with 95% being the downgrade criteria

still leaves me with 288 entries (from 67% TP to 94.7% TP) to make decisions on :)

@RyanM-Regenerateresponse Can you approve a PR for the website repository?

I'm not sure offhand, but I would prefer to let an administrator make the call regarding the details of any public statements.

@Spevacus lol :)

@Spevacus You should see the spam wave ones. The longest of those is 7,460 characters (currently). :) Or the ones I crashed MS with ... more than once, or SD...

@JeffSchaller If you can get rid of the FP, that's great. However, with 660TP, 9 FP, and 1 NAA, that's 670 total (assuming no mixed feedback), so 98.5% TP, which is fine for the blacklist. Ideally, it would be 100% TP, and look-arounds are a reasonable way to improve it, particularly if it looks like the FP are ones which can be prevented in the future. Try, if possible/reasonable, to have the look-arounds be well into the regex, so they are only run rarely to exclude a near match.

@Makyen Holy... Or, well, I should say... unholy

@gparyani I approved it; reminder to roll back once the strike is over

@JeffSchaller also depends on the precise regex; if it's pure cosmetics to remove coincidental FPs like "how do I scrape stuff from israelbigmarket.com"; then just live with having a few FPs

@tripleee github.com/Charcoal-SE/charcoal-se.github.io/commit/…

@tripleee Might I suggest wrapping a <blockquote class="alert"> around that <h1> tag in order to make it more visible?

@CodyGray-onstrike I thought it was plenty visible already, but sure

I don't know, it's only level-1 heading ;-)

But seriously, I thought it blended in somewhat with the content, especially since it was right after the logo banner.

build in progress

not a success )-:

looked okay in preview, but the builder planted the HTML tags verbatim on the page

Oops. The editor sanitized them, I guess.

reverted, build in progress

Gotta say it's weird seeing this room full of human chat messages.

Could someone post, like, 6 lengthy messages about API errors? It'd make it feel more familiar.

Error 404: community trust not found.

Even better than expected <3

May or may not have been ChatGPT generated.

No way to know

Not without being biased against my part of the world.

I wasn't sure, since previously it has not shown a particular aptitude for understanding rhyming. Or being able to count lines in a poem.

Oh, I've seen Jon Clements post several poems written by the AI, and all were pretty solid, with rhyming.

I asked it to generate an API error and it just produced JSON. I said that I need it in plain text but it only generated a short 500 message. Basically "Server error, contact system administrators". If I asked to make it rhyme it said it couldn't. So in the end the prompt was "Write a plain text API error in rhyme".

I mean hypothetically if I were to use ChatGPT that's what I would have done.

:P

Donald Knuth managed to get it to write a (near) sonnet about haikus, with the right rhyming structure

cs.stanford.edu/~knuth/chatGPT20.txt?utm_source=tripleee

(well, "infuse" doesn't really rhyme with "mood" and "imbued" but generally speaking)

@CodyGray-onstrike that should be a 500-series error, don't you think?

?utm_source=tripleee really? Does Knuth pay you for linkbacks?

@tripleee Dunno, as far as I know, "not found" is a 404 error, but I'm no expert. Maybe I should ask a language learning model...

or a large language mentalist, even?

I would imagine Google Bard would be better at writing stuff like that, considering it was originally designed to create unique and creative lyrics and poetry.