Search papers, labs, and topics across Lattice.
Affiliation:
4
2
7
LLM agents trained solely in rule-generated fictional worlds achieve state-of-the-art transfer to real-world search tasks, outperforming agents trained on actual data.
A 5-step sampling schedule can deliver nearly the same quality as a 50-step one, slashing inference costs by 90%.
Ditch the complex diffusion sampling methods: a simple Adam optimizer can give you state-of-the-art image generation and restoration results.
By pausing to "think" with latent diffusion, STAR-LDM achieves superior language understanding, narrative coherence, and controllable generation compared to standard autoregressive models of similar size.
