Search papers, labs, and topics across Lattice.
13
0
16
FlowTracer reveals that optimizing token-level rewards based on attention-induced information flow can dramatically enhance reasoning performance in LLMs.
AURA reveals that understanding implicit user intent can dramatically reduce the number of queries needed while enhancing the relevance of responses.
LLM agents can learn to use tools more efficiently and accurately by explicitly learning when *not* to use them, leading to a 25% increase in tool productivity.
LLMs can convert expensive robot design evaluations into reusable, auditable design principles by distilling search traces into explicit natural-language skill libraries.
Forget specialized architectures: StepAudio 2.5 proves a single audio-language foundation, shaped by RLHF, can dominate ASR, TTS, and real-time dialogue simultaneously.
Current video generation benchmarks miss the forest for the trees: EvalVerse actually measures cinematic quality, not just prompt adherence.
Interactive 3D asset generation can now be driven by functional logic and hierarchical physics, thanks to a new framework that synthesizes simulation-ready assets.
Image-based latent actions are your secret weapon for long-horizon reasoning in VLAs, while action-based latent actions unlock complex motor coordination.
LLMs' own self-judgments, when logically linked to their response features, can significantly improve hallucination detection.
Vision-based tactile signals in the VTOUCH dataset significantly enhance bimanual manipulation capabilities, paving the way for more effective robotic interactions.
Domain-specific reasoning and tool coordination in materials science no longer require massive LLMs: a lightweight, dual-model agent outperforms larger general-purpose models while slashing hardware costs.
A lightweight, RL-trained context curator can match GPT-4o's context management abilities, slashing token consumption by 8x and opening the door to efficient long-horizon LLM agents.
LLM-generated text detection gets a major upgrade: RACE spots the difference between AI as author versus AI as editor, unlocking policy-aligned regulation.
