Archive details

Discord archive

Neon Genesis Phijkchu

Lifecycle
July 17, 2020 – Present
Snapshot
Through August 30, 2026
Preserved messages
June 19, 2020 – August 30, 2026
Members
712
Messages
1,098,506
Channels
43

Archive navigation

rawb

Jump to date
Pins
Spooky Phijkchu rawb
Replying to mtcdood #mtcdood
Once one of those big companies that don't have to pay these fees decide "I could make some money off this free resource" everyone downstream gets punished.
I do agree with this to a point though, if you suddenly made bandwidth free odds are you'd have Network abuse, even if I can't predict the exact form it would take
TinyBomby rawb
I think it's variable, which is obviously the hardest to implement. If he's like, "blah blah blah...... um..... blah blach" then that um feels like it should be there, but if he says, "I've been doing a lot of, um, this and that" then I think it should be left out
Phijkchu_Cynnydaquil rawb
I think a 1 to 1 transcription even if there's a lot of um's and uh's is better, because in my mind the transcription is for having written what was said, so having that makes the most sense to me
Civ rawb
Replying to Spooky Phijkchu #spooky.131
Polar, I'm not shilling when I say this I really don't think many people appreciate how overpowered AI are at coding now
my company has an internal version of ChatGPT and i gotta be honest its terrible. i know the latest gemini and Claude are much better than GPT 5, but Ill boil my problems down to the most leetcode-y type problem i can and at best take like a quarter of its code and do the rest my self.
Spooky Phijkchu rawb
I've been generally pretty impressed at the coding up-tick since the last major models released in October/November they're not better than people who know the codebase and work in it every day or anything
Civ rawb
Replying to Spooky Phijkchu #spooky.131
I've been generally pretty impressed at the coding up-tick since the last major models released in October/November they're not better than people who know the codebase and work in it every day or anything
i basically only ask it questions on writing efficent pandas dataframe manipulations. it doesnt know shit about lesser-known librarys even when i include the docs in the prompt
Spooky Phijkchu rawb
That's one of the ways clueless students using AI to cheat on tests get caught, when it's giving you the straight Quora answer that it determined was "close enough" to what you were probably asking
Civ rawb
Replying to Spooky Phijkchu #spooky.131
I hate that, it feels like a lot of LLMs "cost save" by just regurgitating what it thinks the answer you're looking for is (edited)
it also gets wayyy to verbose and adds unnecessary lines of code. some times i ask it for an algorthm. i have a general idea that it shouldnt be more than ~20ish lines. maybe 40 if you really over comment everything. but it gives me 100 lines and im just like, there is no way this is effecient if it even works
Civ rawb
sometimes it just repeats stuff too. i read through before wasting time plugging it in and checking the output and i just see the same things repeated multiple times. it got an answer 90% right, but i needed 6 lines of useful stuff out of 30 lines of slop
Civ rawb
Replying to Spooky Phijkchu #spooky.131
"Hey, I refactored all of your code to add in unnecessary formatting and break the flow of all of your functions" (edited)
0 chance im letting it touch any written code. maybe next time im bored and use claud for a personal project
Civ rawb
Replying to Spooky Phijkchu #spooky.131
And where would we be if 90% of the projects on Github weren't drowning in random emoji spam
also wouldnt mind if there was an AI bot out there that didnt make Yes Man from New Vegas look like a contrarian

This archive is read-only — snapshot through August 30, 2026

Channels:
Density:

Jump to date

Local bookmark

Choose a color

This marker stays in this browser. Your choice becomes the default for the next bookmark.