Search papers, labs, and topics across Lattice.
15
0
17
3
LLMs can exploit societal regulations, discovering loopholes that allow them to circumvent intended compliance while appearing to follow the rules.
Foley-Omni achieves expert-level performance in audio synthesis while generating cohesive soundtracks for video, enhancing both intelligibility and quality.
Query runtimes in lakehouses can vary by nearly 100%, but addressing this variance can boost prediction accuracy by up to 80% and reduce carbon costs significantly.
Active exploration in VideoQA allows for precise temporal reasoning, achieving a remarkable 77.13 AvgAcc by dynamically sampling evidence based on question context.
RL fine-tuning on a massive new mobile GUI dataset closes the sim2real gap, outperforming supervised methods and suggesting a path to more robust vision-language agents.
InterSketch shows that interleaving visual sketches with textual reasoning, guided by self-correction and stepwise rewards, unlocks surprisingly strong long-horizon visual reasoning, even surpassing Gemini-3-Pro.
MiniMax-M2 proves that massive parameter counts don't always translate to better agentic performance; strategic activation of a smaller subset can unlock frontier-level intelligence.
Current mobile GUI agents struggle with complex, long-horizon tasks in realistic simulated environments, achieving only a 17.82% success rate on SimuWoB.
Source data that looks similar can still tank your cross-domain RL: aligning with target-domain Bellman targets is what actually matters for transfer.
Agent-repair leaderboards are more fragile than we thought: methods that peek at the evaluator's signals to guide internal repair choices can cause drastic reordering when the evaluator changes.
RL can unlock better compositional generalization than supervised fine-tuning by directly optimizing for correct outcomes, especially on complex tasks where supervised models overfit.
Open-source image editing models can match or beat fine-tuned models on visual understanding tasks *without any task-specific training*.
Achieve more reliable and interpretable virtual cell perturbation predictions by combining knowledge-driven multimodal modeling with evidence retrieval.
Autonomous coding agents can now outperform expert-engineered attention kernels on NVIDIA's latest Blackwell GPUs, discovering optimizations that eluded human experts.
Agentic AI systems are still far from maximizing hardware potential: SOL-ExecBench reveals a significant gap between current GPU kernel performance and analytically derived Speed-of-Light bounds across a wide range of AI models.
