Search papers, labs, and topics across Lattice.
6
1
8
6
Vision-language models can now achieve remarkable out-of-distribution accuracy by effectively leveraging self-reflection to correct errors in multimodal reasoning.
Randomized YaRN boosts long-context reasoning performance by exposing models to out-of-distribution positional encodings, yielding impressive gains at extreme lengths.
GENIE reveals that traditional metrics fail to capture the nuanced dimensions of novelty, offering a sharper lens for evaluating LLM creativity.
LLMs struggle to generate diverse and specific connections between concepts, even with high token budgets and "thinking" prompts, revealing a gap in creative associative reasoning.
LLMs can learn to make better decisions in complex environments by explicitly reasoning about the cost of exploration, leading to more efficient information gathering and problem-solving.
Forget fixed margins in RLHF: modeling the *strength* of human preferences with "preference-over-preference" learning boosts both discriminative accuracy and generative quality.
