Kiran

2.6K posts

Kiran banner
Kiran

Kiran

@kirancodes

Research Scientist in NYC I work in Programming Languages research, on formal verification. British South Asian diaspora~

US Katılım Kasım 2011
193 Takip Edilen1.5K Takipçiler
Sabitlenmiş Tweet
Kiran
Kiran@kirancodes·
Happy thanksgiving
Kiran tweet media
English
2
2
20
22.7K
Kiran retweetledi
robodu
robodu@rxbodu·
having nothing to contribute to the current parental trauma conversation because you dissociated through most of it and can't remember that period of your life anymore.
English
40
580
2.7K
48K
Kiran retweetledi
cyn🌸
cyn🌸@cynugwudike·
“and she saw me bleeding” a piece from my “does my mother love me?” collection.
cyn🌸 tweet media
English
475
4.7K
36.2K
29M
Kiran retweetledi
Mayor Zohran Kwame Mamdani
Every confusing city website and dead-end government portal makes New Yorkers trust their leaders less and less. It's time for City government to enter the 21st century. We’re investing in five @NYCOfficeofTech PIT crews dedicated to modernizing New York City’s web tools and creating new, easy-to-use online resources. If you’re a software engineer, designer, or technologist who wants to help make City government work better and faster, apply to join our PIT crews at nyc.gov/pitcrew.
Mayor Zohran Kwame Mamdani tweet media
Mayor Zohran Kwame Mamdani tweet media
English
327
861
9.7K
649K
Kiran
Kiran@kirancodes·
@subcountability oh yeah detexify is such a life-saver! I used it all the time during my PhD! It's crazy how much more we can now do with the frontier models! Science is amazing!
English
0
0
1
10
nk
nk@subcountability·
@kirancodes One of the longest online ML tools, detexify from 2009, let's you find symbols by drawing them in a paint like style detexify.kirelabs.org
English
1
0
0
42
Kiran
Kiran@kirancodes·
never writing latex by hand ever again. You are not ready.
English
32
15
259
34.5K
Kiran
Kiran@kirancodes·
@rbarcrosspbar The latency isn't processing time, its artificially there for denouncing and as part of the design. Ive found that idiots are often reticent to think about what they've written before moving forward and would find pauses in their stream of slop an issue. Maybe this isn't for you
English
0
0
0
104
Kiran
Kiran@kirancodes·
@fcoder then don't use it????
English
0
0
0
123
fcoder
fcoder@fcoder·
@kirancodes It seems insanely slow and unreliable to me. How can I use it during the class, if I have to wait and revalidate results constantly?
English
1
0
0
275
Kiran
Kiran@kirancodes·
@yaga_simha right exactly?? the models are surprisingly good at this. chuck it on an eink display and just cut down the iteration loop
English
0
0
0
154
Yaga Simha
Yaga Simha@yaga_simha·
@kirancodes I've already been exploiting Gemini to turn my handwritten formulae into LaTeX!
English
1
0
1
226
Kiran
Kiran@kirancodes·
@joseph_h_garvin ooh good point, having some audio input would also be pretty rad; might try throwing that together too
English
0
0
1
69
Joseph Garvin
Joseph Garvin@joseph_h_garvin·
@kirancodes Doing the x like you did is cool but instead of writing the instructions should use voice
English
1
0
0
166
Kiran
Kiran@kirancodes·
@xLexiLiesx I tried a bunch and for my handwriting light on ocr worked the best
English
1
0
0
23
LexiLies
LexiLies@xLexiLiesx·
@kirancodes Do you know of a good model for that? I was thinking a diffusion for the block transcription to a scout model for context and prose consistency confirmation before asking a small 4-8B model to then transcribe it and spell check me. Im probably over complicating it... lol
English
1
0
0
27
Kiran
Kiran@kirancodes·
@xLexiLiesx It relies entirely on the image to text transcription model for converting your handwriting to text. So if you use a model that handles cursive well then it would work.
English
1
0
0
265
Kiran
Kiran@kirancodes·
@taziksh Nope, but i have heard good things!
English
0
0
0
161