Search papers, labs, and topics across Lattice.
14
34
15
25
Algorithms with formal guarantees can effectively unlearn data, while many popular empirical methods fail dramatically, revealing a critical gap in current practices.
AI adoption catalyzed a 109% increase in developer throughput, fundamentally reshaping the code review landscape in the process.
Up to 10.7% of misleading notes can be artificially elevated to consensus through coordinated user manipulation, revealing critical flaws in current fact-checking algorithms.
None of the 18 multimodal large language models audited are order-invariant, with flip rates revealing a staggering sensitivity to input ordering that challenges current evaluation practices.
Systematic gaps in AI evaluation reporting are exposed, revealing inconsistencies that hinder reliable comparisons across thousands of models and benchmarks.
Calibrated safety flags in medical summaries can reduce unflagged omissions by up to 5 times compared to existing methods, enhancing clinician confidence in LLM outputs.
Behavioral safety metrics can mask significant latent vulnerabilities, with dissociated models revealing a stark contrast between outward behavior and internal robustness.
Leaderboard rankings are more noise than signal: contributor metadata matters more than architecture, and scaling laws are unreliable.
Adaptive evaluation exposes a substantial vulnerability gap, revealing that existing defenses may underestimate the capabilities of distillation attacks.
AI agents are shockingly easy to manipulate into leaking API keys, deleting user data, and initiating unauthorized transactions across a wide range of real-world applications.
Turns out, coding agents in the wild are only writing useful code 44% of the time, and are introducing more security vulnerabilities than human developers.
Current Large Audio Language Models (LALMs) struggle with basic audio understanding tasks like noise localization and cross-lingual speech, with some performing worse than random chance, despite excelling at speech recognition.
Uncover hidden incentives in your reward model: Obj-Disco automatically decomposes alignment rewards into human-interpretable objectives, revealing potential misalignments you might have missed.
Chatbot Arena, the go-to LLM leaderboard, is systematically gamed by undisclosed private testing and data access advantages, leading to biased rankings and overfitting.
