Training a GPT on massive amounts of fMRI data

BRAINLM: A FOUNDATION MODEL FOR BRAIN ACTIVITY RECORDINGS

My Twitter, LinkedIn, Discord, Patreon, consultation booking page, etc:

Timestamps:
00:00 intro
01:15 timestamp
02:34 timestamp
04:37 timestamp
05:57 timestamp
07:06 timestamp
07:15 timestamp
12:45 timestamp
16:34 timestamp
17:43 timestamp

[erikhartman]

Is r2 not the coefficient of determination? If so it would be defined in the range (-inf, 1], and -1 is worse than 0. Just an FYI as I know you're on your ML journey!

[webgpu]

hmm, this channel seems pretty cool, actually. Dude, thank you for digesting these papers for us. Greatly appreciated! 👍👍

[tornyu]

Cool research direction!

When you were describing *G* I was thinking, ha! that sounds a lot like an adversarial attack from an alignment paper.

[acasualviewer]

I hadn't seen spacial embeddings before. That's interesting.

[kylev.]

D waves are great. You don’t need an fMRI, you can use an EEG

[nofreewill]

Next up: researchers are scanning for passwords in fMRI scans.

[cammccauley]

I wonder if data from neuralink could be substituted for the fMRI data.