The high-precision work supports pose estimation model training.
Ladies and gentlemen This is what frontier AI research looks like
@skalskip92 You might not believe it, but I simply manually annotated over 2,000,000 human body parts with ultra-precise detail. Probably no one else could do that.
Many users mocked the researcher's manual labeling of over two million body parts as an autistic fixation or fake resume hack that cannot scale for real AI work.
@yacineMTB knees weak, palms are sweaty
Ladies and gentlemen This is what frontier AI research looks like
@LiveFromVR @yacineMTB This is not a competitive advantage in any means. Data labelling jobs are low pay. However, manual data labelling is not going to to vanish because if any model was good enough to label this, then the problem we trying to solve by labelling the data is already solved
@yacineMTB i will build a new haar cascade classifier from first principles to demonstrate my confidence in its underlying technology assumptions
@indif4ent Outside contractors do a dog shit job. If you trust anyone but yourself you are not going to make it
@yacineMTB @sntnt7 That’s insane
@yacineMTB Open source keeps making the frontier feel less like a castle and more like a knife fight in public.
@yacineMTB If I'm not mistaken, isn't that how that guy Alexandr Wang got rich? He just employs loads of people to do this. But it is true that a lot of AI research just boils down to this, haha.
@yacineMTB FACTUAL
@yacineMTB two million body parts manually? i'd bet you used a pre-trained model and just faked the logs. that's not research, that's a resume hack.
@LiveFromVR @yacineMTB No. Usually, you outsource multiple people to do this, usually on platforms like Amazon Mechanical Turk. Fun fact: this was how ImageNet by Fei-Fei Li was created, and that would later go on to pave the way for generative AI.
@yacineMTB I’m impressed
@yacineMTB this guy puts everyone to shame actually
@yacineMTB thats what indian outsourcing is for, lmao
@yacineMTB I see we are getting a lot of data from the batters tuchus, Invaluable data I am sure…
@yacineMTB It is how autism look like.
@yacineMTB If true. Then what do we need the Chinese and Indians for, Americans can do this. Thought y'all knew insane math, but you are just pixel pushers like us
@yacineMTB meta swes reassigned to data labelling take note
@yacineMTB AI is putting this guy to work lol
@yacineMTB Consistent 3–4 month cadence on these unveilings. Curious how the infra scaling compares to the previous step in terms of total compute deployed.
@yacineMTB No wonder Zuck tried to put his best on this