Search papers, labs, and topics across Lattice.
7
0
8
58
Scaling linear RNNs with Sparse Delta Memory leads to dramatic gains in long-context recall without increasing computational overhead.
DecompRL enables LLMs to solve complex problems by breaking them down into manageable sub-tasks, achieving a 50x reduction in GPU costs while enhancing solution diversity.
Extrapolating between code-generating RL agents trained on different unit test coverages unlocks better correctness-efficiency trade-offs than any single agent alone.
LLMs can't rebuild software from scratch, even for widely used programs like FFmpeg and SQLite, revealing a critical gap in their ability to make high-level software architecture decisions.
Agentic coding gets a serious boost: distilling and reusing rollout trajectories lets Claude-4.5-Opus jump from 70.9% to 77.6% on SWE-Bench Verified.
LLMs can now automatically verify imperative code at scale, achieving state-of-the-art results on challenging verification benchmarks and paving the way for large-scale verified code datasets.
LLMs can now emulate debuggers, stepping through code and setting breakpoints, opening the door to more interactive and controllable neural program execution.
