Multi-step planning is a challenge for LLMs. Here, the authors introduce a brain-inspired Modular Agentic Planner that decomposes planning into specialized LLM modules, improving performance across tasks and highlighting the value of cognitive neuroscience for LLM design.

When faced with many options to choose from, humans typically need to explore the utility of new choice options. People with an autism diagnosis or elevated autism traits are thought to avoid exploring such unknown options, but it remains unclear how autism affects exploration in decision spaces with many options. In a large online sample (N = 588), we investigated the impact of autism diagnosis or elevated autism traits on exploration behavior during value-based decision-making in vast decision spaces. We used a 121-armed bandit with spatially correlated choice options, and a dedicated computational model to disentangle generalization, uncertainty-guided exploration, and random exploration strategies. Our findings show that participants with a self-reported autism diagnosis were less likely to explore novel choice options and more likely to exploit known high-value options. Computational modeling suggests they engaged in less uncertainty-driven exploration but exhibited equal random exploration and generalization strategies. Interestingly, among non-diagnosed participants, people with elevated autism traits did not explore less. This study relies on self-reported autism diagnoses and trait measures collected online. This may limit the generalizability of the findings to clinically verified or more diverse autism populations. Our findings highlight important differences in exploration strategies between clinical and subclinical populations and emphasize the importance of cognitive modeling and using vast decision spaces to better understand autism.

Scientific Reports - Asian elephants (Elephas maximus) recognise human visual attention from body and face orientation

APPLICATION INSTRUCTIONS: CURRENT PENN STATE EMPLOYEE (faculty, staff, technical service, or student), please login to Workday to complete the internal application process. Please do not apply here, a...

In a landmark study, OpenAI researchers reveal that large language models will always produce plausible but false outputs, even with perfect data, due to fundamental statistical and computational limits.

ALT: a man in a hat is reading a book in a library .