dito

making sense of computers one yap at a time

[dito @rayendito]

@meetapersecond anjay

[Daniel W. @ghostinthecable]

@cheddar420yolo I am everyone's weird coworker

[Amazon Web Services @awscloud]

More AI-generated code doesn't make your team faster. It might actually slow you down.

[dito @rayendito]

cite TimpaTeks pls @google we have a pretty cool way to edit text in-place kinda lol

https://t.co/1TgjDcrRRD

[Alham Fikri Aji @AlhamFikri]

🚨Introducing TimpaTeks!

We’re exploring activation steering in diffusion models, allowing for in-place editing to transform text into new concepts.

Amazing work led by @rayendito

Check out the paper: https://t.co/BZogkHjMVo
Github: https://t.co/6hVlcHtCOW

[dito @rayendito]

yyyyooooooooooo

https://t.co/mvAETb2UaW

this is unironically more exciting than the fable release for me lmao

[dito @rayendito]

EVERYTHING THAT HAS BREATH PRAISE THE LORD

[dito @rayendito]

@IkhlasulHanif0 waduh

[Hanif | AI NOT FOR PRODUCTIVITY @IkhlasulHanif0]

🚨 New Paper Alert! 🚨
How does AI bias manifest in an incredibly diverse nation like Indonesia? 🇮🇩

If you think AI bias is complicated in the US, try looking at Indonesia.

With hundreds of ethnicities/languages and a multi-party democracy (yes we have like 10+ parties), i.e LLMs heavily favor massive majority groups (like Sundanese) while completely sidelining marginalized ones (like Korowai).

paper link: https://t.co/2sU0NT4UW0

[dito @rayendito]

@erla_ndpg 😍

[Edd @erla_ndpg]

Diffusion + Interp. My favourite combination 😍

You guys should check their work!!!

[dito @rayendito]

@faizkrisnadi @AlhamFikri thanks! i gotta admit it's a bit messy though, but refactors are in order heheh

[dito @rayendito]

@IkhlasulHanif0 @AlhamFikri

[dito @rayendito]

@bandungsaeki rapi dikit jadi hoax

[dito @rayendito]

You ever wanted to change a piece of text (positive to negative movie review, sentence abt cats to dogs, etc) without changing the structure of the sentence too much? our project, TimpaTeks🗿 lets you do that 😭😱🤯

TimpaTeks🗿 is based on diffusion LMs and activation steering. Basically, to change the sentence automatically, it selects (samples) tokens that needed to be replaced, then it goes through the usual DLM generation pipeline with steer vectors injected.

I think the results are pretty neat and the text is still coherent (gif is an actual run result) pls check it out guys

github: https://t.co/pNW6vOYyLs
arxiv: https://t.co/0yTiHfNOXZ

[dito @rayendito]

@_larpGPT that may or may not have been one of the main motivations

[Alham Fikri Aji @AlhamFikri]

🚨Introducing TimpaTeks!

We’re exploring activation steering in diffusion models, allowing for in-place editing to transform text into new concepts.

Amazing work led by @rayendito

Check out the paper: https://t.co/BZogkHjMVo
Github: https://t.co/6hVlcHtCOW

[dito @rayendito]

@AlhamFikri alternative title

[dito @rayendito]

@IkhlasulHanif0 @AlhamFikri i forgor, fadli's twitter is @epiceuwiw

[dito @rayendito]

this paper came to be with the help and inspiration from my cool friends @IkhlasulHanif0 @ Fadli Aulawi @ Ahmed Elshabrawy @AlhamFikri

... and the TimpaTeks memes (literal translation: TextOverwrite memes)

[dito @rayendito]

@ImperialistsL what im most excited about tbh

[dito @rayendito]

@zmkzmkz i've been ksabar waiting to tweet a paper titled ksabar