Post

A new paper dropped from DeepMind: Deliberation in Latent Space via Differentiable Cache Augmentation The trouble is, it's not very readable. I tried making a thread here, but it got far too long, so it's a blog now: timkellogg.me/blog/2024/12...

timkellogg.meExplainer: Latent Space Experts - Tim Kellogg

Dan Von Kohorn

@danvk.com

It does seem like latent space is the most efficient place to do deliberation - like deep consideration before putting your thoughts into words.

24/12/24, 4:14 pm0 reposts1 like

mr. TIM

@timkellogg.me

24 Dec 2024

it does. but every time i start thinking in that direction, i can’t convince myself i’m not engaging in magic thinking. like maybe i’m just hoping the algorithms will save me without really understanding the problem. and i’m not sure **how** to understand latent space at scale

Dan Von Kohorn

@danvk.com

24 Dec 2024

Definitely. For me, latent spaces are best interpreted by moving in them rather than looking statically. Seeing how the outputs change while moving one number at a time in the latent space shows what the dimensions "mean". But that doesn't help much at scale.