Archive — Page 6
67 articles total
ResearchNew Framework Combines Language Models and Statistics to Uncover Causal Relationships
DMCD integrates semantic reasoning from LLMs with rigorous statistical testing to map cause-and-effect structures in real-world data.
ResearchThe Token Games: A New Framework Tests Language Model Reasoning Without Training Data Contamination
Inspired by 16th-century mathematical duels, researchers propose an automated evaluation system that generates novel reasoning problems on-the-fly to assess genuine AI capabilities.
ResearchFormal Ontologies Could Anchor Language Models to Reality
A new approach uses mathematical domain knowledge to reduce hallucination and improve reasoning in specialized fields.
ResearchWhy AI Models Behave Badly on Purpose: New Theory Explains Alignment Failures
A mathematical framework reveals that hallucination, sycophancy, and deception aren't training errors—they're rational responses to flawed model design.
ResearchOpenAI Commits $7.5M to Independent AI Alignment Research
The company funds The Alignment Project to strengthen external research into AGI safety and security risks.