hallucinations
4 mentions across all digests
Hallucinations are instances where large language models generate factually incorrect or fabricated outputs that appear plausible, an active research area with emerging benchmarks, causal graph frameworks, and multi-agent consensus techniques aimed at mitigation.
KARL: Mitigating Hallucinations in LLMs via Knowledge-Boundary-Aware Reinforcement Learning
Researchers propose KARL, which combines reinforcement learning with knowledge-boundary awareness to teach LLMs when to decline low-confidence responses, directly tackling the persistent hallucination problem by aligning model outputs with actual training data coverage.
Unmasking Hallucinations: A Causal Graph-Attention Perspective on Factual Reliability in Large Language Models
Causal graph-attention framework reveals how attention mechanisms contribute to LLM hallucinations, enabling more precise diagnosis of factual errors.
Hallucination Basins: A Dynamic Framework for Understanding and Controlling LLM Hallucinations
Researchers propose Hallucination Basins, a dynamic framework that maps and controls confabulation patterns in LLMs, offering a systematic approach to suppress a fundamental reliability failure mode.
The Hallucinations Leaderboard, an Open Effort to Measure Hallucinations in Large Language Models