Maximilian Kasy (@maxkasy) 's Twitter Profile
Maximilian Kasy

@maxkasy

Econ prof @OxfordEconDept, econometrics, inequality, machine learning, politics. Mastodon: @[email protected] bsky.app/profile/maxkas…

ID: 2294500723

linkhttps://maxkasy.github.io/home/ calendar_today16-01-2014 14:04:24

2,2K Tweet

10,10K Followers

893 Following

Maximilian Kasy (@maxkasy) 's Twitter Profile Photo

Great interview with Meredith Whittaker on Signal, #privacy, #ai, and the #surveillance business model: wired.com/story/meredith…

Maximilian Kasy (@maxkasy) 's Twitter Profile Photo

Starting Monday: Department of Economics, Oxford summer school, incl foundations of #ML for #economics: maxkasy.github.io/home/ML_Oxford… New this year: 1) Intro to supervised learning using #python: maxkasy.github.io/home/files/tea… 2) Transformers for language processing maxkasy.github.io/home/files/tea…

Maximilian Kasy (@maxkasy) 's Twitter Profile Photo

What are the returns to data size D, compute C, model size N in AI? This paper estimates "production functions" - dependence of predictive loss on these params (ht Friedrich Geiecke): arxiv.org/abs/2001.08361 Lots of interesting ramifications: 1) For learning theory: 1/n

What are the returns to data size D, compute C, model size N in AI?
This paper estimates "production functions" - dependence of predictive loss on these params (ht Friedrich Geiecke):
arxiv.org/abs/2001.08361

Lots of interesting ramifications:
1) For learning theory:
1/n
Maximilian Kasy (@maxkasy) 's Twitter Profile Photo

What are the returns to data size D, compute C, model size N in AI? This paper estimates "production functions" - dependence of predictive loss on these params (ht Friedrich Geiecke): arxiv.org/abs/2001.08361 Lots of interesting ramifications: 1) For learning theory: 1/n

What are the returns to data size D, compute C, model size N in AI?
This paper estimates "production functions" - dependence of predictive loss on these params (ht Friedrich Geiecke):
arxiv.org/abs/2001.08361

Lots of interesting ramifications:
1) For learning theory:
1/n
Maximilian Kasy (@maxkasy) 's Twitter Profile Photo

Reading rec: "Ideology: An Introduction" by Terry Eagleton versobooks.com/en-gb/products… A tour the force through 200 years of theorizing about ideology, and the relation between ideas, power, truth, and social change.

Maximilian Kasy (@maxkasy) 's Twitter Profile Photo

Just found out about Altair, a #dataviz library for #python with a very clean syntax based on the "grammar of graphics:" altair-viz.github.io Intuitive for ggplot users! Just substitute: geom -> mark aesthetic -> encode