HOLY FUCKING SHIT!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
IT HAPPENED!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
FIRST INTERNET MONEY!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
IM SO HAPPY!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
ai text has tells: dashes, bold, bullets, "great question!". the english airplane manuals use since 1986 is almost the opposite
the rulebook is simplified technical english (asd-ste100). example, the simpleenglish skill turns it into rules for your ai:
- first sentence answers the question
- no em-dashes
- no "great question!" or "hope this helps!"
same claude model in the readme: before, "leveraging", "robust", "seamlessly", "crucial". after, two short sentences
the catch: it reads flat. fine for docs, keep your own voice for the rest :)
comment simple and i'll send you the link. which tool should i try next?
4
2
169
sources: the simpleenglish readme on github (aminblg/simpleenglish, v2.1.2) and the official asd-ste100 site, which says ste became a requirement for commercial aviation manuals in 1986. author: @sotch_desu
2
51
a pngtuber is just two pictures: mouth closed, mouth open. the player swaps them with the voice
example, livliv player (free, apache-2.0, runs in the browser):
- load your two pngs with PngBody
- play an audio file with createSpeech, the mouth follows it
- it adds the sway, tilt and breathing on its own
most of its ready-made characters are fine for creators, but companies have to ask first. or bring your own drawing. this is the pico doing it :)
comment livliv and i'll send you the link. which tool should i try next?
94
til the thumbs up/down in chatgpt can let openai use that whole conversation for training, even with training switched off
- openai's own help page: rating a reply means "the entire conversation associated with that feedback" may be used
- on free, plus and pro the training switch is on by default anyway
- fix: settings > data controls > "improve the model for everyone" off. chats you don't rate stay out, your history stays
- and on a private chat, just don't rate it :)
2
2
147
source: arena.ai's blind-vote leaderboards (creative writing, webdev, text-to-image, image edit, search, text-to-video, text), boards updated nov 6, read nov 7
21
gave a local 8b model (gui-owl 1.5) my android phone today
told it "open the clock, don't start a timer"
it started a 10 min timer :)
so now:
- plain code reads the screen + taps
- the model only answers yes/no
- on my gpu, ~1s an answer, $0
small models = judges, not drivers
1
84











