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...

interesting, i’ve heard of CAVs, but they always seemed more of a gray box approach. it seems like you could train a model to translate an embedding back into text. doesn’t seem like it would be too hard. also seems inevitable if we go all in on latent space

I believe the latent space almost always means that there's potential for more accessible information than just text with the space being continuously differentiable. I think of early days variational autoencoders that could be used to generate beyond training data

But translating embedding back to text seems to me not necessarily straightforward. Where I've seen success is when comparisons are done with a 'well-understood' embedding. I can see conversion to text being difficult in a similar way that there doesn't exist perfect translation between languages