rdgao.bsky.social
computational neuroscience, machine learning, occasionally cognitive science | inverting brain circuits towards translational neuroscience.
🌊 www.rdgao.com
91 posts
876 followers
346 following
Regular Contributor
Active Commenter
comment in response to
post
plot twist: it synchronize wiggles with gamma oscillation
comment in response to
post
I love the "but the worm is different tho" finale.
comment in response to
post
comment in response to
post
awww thanks 🥰 but I think it's fair to lump me with LLM considering how I have the same 4 or 5 auto-generated rants
comment in response to
post
No disagreements there!
comment in response to
post
yes you
comment in response to
post
didn't somebody get mad at me a few years ago for asking what working memory is?
comment in response to
post
XJ Wangs early (~2002) decision making stuff? Not trajectory explicitly (I think?) but fixed points in state space.
And definitely Walter Freeman III
comment in response to
post
I think we can say something more broad:
most of neuroscience has been and will be useless until we can figure out how to infer signals we don't have from signals we do have, foundation model or not.
comment in response to
post
somebody in silicon valley is salivating over this tweet rn
comment in response to
post
oh fuck that's brilliant
comment in response to
post
absolute banger
comment in response to
post
"We hereby uncover the fundamental principle of deep learning..."
comment in response to
post
what about just "the methlab" tho
comment in response to
post
wow okay evidently my original tweet was made for you not your postdoc
comment in response to
post
you still have time for eating and drinking in this dumpster fire?
comment in response to
post
That makes three of us
comment in response to
post
this is really beautiful stuff, and a very nice explainer thread
comment in response to
post
what the absolute fuck
comment in response to
post
no not THAT good, more like a Lou Will
comment in response to
post
6th man of the year
comment in response to
post
🥹
comment in response to
post
looking forward to many more Twitter / BS debates on why oscillations and LFPs don't ~do~ anything!
www.cell.com/trends/cogni...
comment in response to
post
so 6 months until "is maxpooling epiphenominal?"
comment in response to
post
so you're the Grinch or what
comment in response to
post
🔥🔥
comment in response to
post
you'll be hearing from god in a few weeks
comment in response to
post
but you can say the same about many other architectural details in practice? Without depth, skip connections, tied convolutions, etc. it doesn't work.
As far as I know, nobody has gotten a 1 layer relu network to do AI stuff? What's the "right" abstraction level to make this argument sensible?
comment in response to
post
damn so I have to convert it myself??
comment in response to
post
how do I convert this:
rdgao.github.io/blog/2024/08...
comment in response to
post
^ this. Better yet, matrix multiply.
comment in response to
post
Feels like that ship has long sailed with no anchors
comment in response to
post
arxiv.org/abs/2407.08751