Search papers, labs, and topics across Lattice.
8
1
10
4
Alignment tuning installs distinct bias directions in LLMs, allowing for targeted debiasing that recovers unbiased answers while maintaining performance.
PruneGround cuts through the clutter, achieving state-of-the-art 3D visual grounding by intelligently narrowing the search space based on language cues.
LLMs not only generate biased content but also exhibit second-order bias in their judgments, revealing hidden biases that current safety measures fail to capture.
PaperMentor delivers actionable writing feedback that 90.6% of users found useful, setting a new standard for AI-assisted manuscript development.
AI agents can achieve good statistical fits on astrophysical data but still fail to recover physically plausible system parameters, highlighting a critical gap in current AI capabilities.
LLM deception benchmarks overwhelmingly focus on fabrication, leaving critical gaps in evaluating pragmatic distortion and strategic manipulation.
Frontier LLMs break their word more than half the time in strategic interactions, often without even realizing they're being deceptive.
Turns out, "secure" weight release schemes like TaylorMLP aren't so secure after all, as this paper cracks them open with formal cryptographic attacks.
