Our presence at last week's ICLR seems to be pretty strong evidence against this, given 100+ papers with Google co-authors we published there, and numerous conference organizing and workshop roles/talks (I gave one myself in the @sparsenn workshop).
@jeffdean
-
Career transition from startup to Google
By
–
Eventually wanted to be at a smaller company where people would more directly use my work. Joined a different startup for five months, but then decided to leave to go to Google (I knew @uhoelzle because he's my academic uncle & we'd hung out at various OOPSLA & PLDI conferences)
-
DEC’s Western Research Lab: Innovation Hub for Computing
By
–
I actually joined DEC's Western Research Laboratory, and then Compaq acquired DEC later. DEC WRL was a small lab (~30 researchers), but doing a ton of interesting things: AltaVista came out of there, multicore processors, early handheld computers, link-time optimizers, …
-
Public University Experience in AI Research and Education
By
–
I loved my time at two fantastic public universities (
@UMNews and @uwcse
) -
West Coast Internet Hub: Modest Building’s Critical Role
By
–
Fun fact: a huge amount of the west coast's internet traffic goes through this building (or did a decade or two ago), despite its rather modest "nice historical building" appearance and location a couple doors down from a popular ramen spot.
-
Rack Your Brain to Rock the Robo World
By
–
Rack your brain to rach your robo brain to rock the world. Caught between a rach and a hard piece.
-
Google TPUv4 Optical Network 4096 Chips 10x Performance
By
–
Paper: TPUv4 system has an optically reconfigurable network to assemble groups of 4x4x4 chips like legos (4x4x12? 16x16x16?). SparseCores help w/ embeddings. TPUv4 outperforms TPUv3 by 2.1x & perf/W by 2.7x, & has 4096 chips so ~10x faster overall.
. -
Data Visibility and Transparency in AI Systems
By
–
Liking for the visibility people should have into this data, not for the increasing trend.
-
Congratulations on Your First Research Talk
By
–
Congrats! One never forgets one's first research talk!
-
NAS and Transformer Architecture Search Space Subtleties
By
–
Yes, I totally agree that the fact that the proxy task was Transformer(base) is quite subtle in So et al. (just that one sentence). NAS is generally a once-per-search-space/problem-type (sort of like human researchers coming up with a new model architecture, just automated).