Good Data Ideas Are Often Simple And Seemingly Dumb

@code_star some of the TDA ideas that work are pretty cool right?

@code_star Number one professional red flag for my entire career has been when I think I’ve outsmarted everyone else.

I've been working with data for so long. Often, it feels like there is nothing new under the sun. Just applying the basics, doing it on a bigger scale, and being insanely paranoid.

Occasionally, though, you will come across an idea so brilliant and simple that you are just blown away that it hasn't already been done in public before.

@code_star +1 every cool data idea ive had hasnt worked

@code_star Took some time off to hone in the vague posting. You absolutely love to see it

@dtcb Shit, you're right. I need to wean myself off it. 😅

@mkurman88 Good point

@code_star rm -fr data/*

no data, no problem

@code_star ok ok, what chinese paper did you read now? :P

@code_star Nah I’m always on board for Arcee hype 😂

@code_star my mind is full of great ideas, don't hesitate to dm if you have any similar problems :D

@aspergtame @code_star GUI grounding datasets are noisy. i will filter them to only ones that seem really correct and helpful that dont have too few objects and dont have too many. result: doesnt matter.

@code_star robotics same way. cool motors fail constantly. the 30-year-old one still works.

@code_star the smartest sounding idea in the room is usually the one that falls apart in the first join

ngl the dumb obvious ones are always the ones that actually ship

@andersonbcdefg @code_star There’s 2 ways to infer that :)

@code_star had this with robot datasets. tight filtering = perfect training, fails day one on a real job site.