Paper Reading | Latent Thought Models
LTM turns extra inference-time compute into explicit per-example posterior optimization over latent thought vectors, rather than more token-space decoding.
LTM turns extra inference-time compute into explicit per-example posterior optimization over latent thought vectors, rather than more token-space decoding.