OmniDrop: Efficient Token Pruning for Omni-modal LLMs
Discover OmniDrop, a novel layer-wise token pruning method that boosts efficiency and accuracy in omni-modal large language models with query-guidance.
InsightReplay: Boost AI Multistep Reasoning Accuracy
Discover InsightReplay, a stateful reasoning method that improves AI multistep reasoning by preserving critical insights for better accuracy.
Intelligence Impact Quotient: Measuring AI’s Organizational Value
Discover the Intelligence Impact Quotient (IIQ), a new metric to measure AI integration and impact in organizations for better decision-making.
Optimizing Prompting Policies for Multi-step Reasoning in LLMs
Discover how iterative distillation and RL improve prompting policies for multi-step reasoning and tool-use in black-box large language models.
Synthesizing POMDP Policies via Sampling and Model-Checking
Discover a novel framework combining sampling and model-checking to synthesize scalable, formally verified POMDP policies for reliable AI decision-making.