Capability Elicitation vs Creation in Post-Training AI Models
Explore how post-training methods distinguish capability elicitation from creation in AI models using a free-energy perspective for enhanced performance.
AI Embeddings for Capturing Preferences in Decisions
Discover how AI embeddings focusing on preferences improve collective decision-making beyond semantic analysis in online voting and deliberation.
Auto-Rubric Reward: Enhancing Multimodal Generative Models
Discover how Auto-Rubric as Reward improves multimodal AI alignment by converting implicit preferences into explicit, reliable evaluation criteria.
Spatial Priming Boosts LLM Accuracy in Chart Data Extraction
Discover how spatial priming with grid overlays improves LLM accuracy in extracting data from scientific charts, outperforming semantic prompting methods.
Reliability in Vision-Language Models: Study of Attention & Causality
Explore how attention, hidden states, and causal circuits impact reliability in vision-language models for improved AI performance and trustworthiness.