Search papers, labs, and topics across Lattice.
6
0
9
LOGOS shows that a single generative model can outperform specialized systems across multiple scientific domains, challenging the need for separate technical stacks in AI for Science.
FlashCP achieves up to 1.63x faster training for large language models by eliminating redundant communication and optimizing workload balance.
Crafter revolutionizes scientific figure generation by enabling multi-type outputs and local editability, outperforming existing systems across diverse benchmarks.
RLVR's reasoning boost isn't just about more data – it's a Goldilocks problem: too easy and LLMs skip the reasoning, too hard and they break down, but just right and they become reasoning powerhouses.
Current VIP identification methods miss the forest for the trees, leading to "Temporal Importance Shift"—but a new model leveraging spatio-temporal cues and rationale generation closes the gap.
Forget complex regularizations – stable sequential knowledge editing in LLMs boils down to correctly managing accumulated constraints, unlocking simpler and more reliable updates.
