Research
Hidden Anchors in Multi-Agent LLM Deliberation
Researchers modeling multi-agent LLM deliberation as a closed-loop dynamical system have identified what they call 'hidden anchors'—internal belief states that persistently pull each agent's expressed opinion regardless of group influence, and that can be recovered from the deliberation record alone. Crucially, these anchors allow agent confidence in a correct answer to exceed the range set by any agent's starting position, a behavior that standard consensus models forbid. The finding has practical implications for understanding when multi-agent reasoning systems are genuinely reasoning versus socially converging, and for auditing the reliability of deliberative AI pipelines.
Read full story at cs.AI updates on arXiv.org →V: · A: · D:
Related
Research
Reinforcement Learning Towards Broadly and Persistently Beneficial Models
Researchers have published findings suggesting that reinforcement learning on carefully constructed datasets of benefici...
Research
Commemorating 70 Years of Artificial Intelligence
IEEE Spectrum marks seventy years since the Dartmouth workshop formally named artificial intelligence as a field, offeri...
Research
Diffusion Language Models: An Experimental Analysis
Researchers present a systematic evaluation of eight diffusion language models across eight benchmarks covering reasonin...