Tag
This paper investigates whether human cognitive diversity in prediction markets can be extracted as behavioral profiles and injected into LLM agents. Extraction partially works with stable profiles, but prompt-level injection fails to meaningfully transfer diversity, suggesting the need for deeper model interventions.
Njord is a probabilistic graph neural network for ensemble ocean forecasting that provides uncertainty estimates and achieves state-of-the-art performance on global and regional benchmarks, improving surface temperature prediction.