Think-Probe-Respond Increases Research Idea Novelty Judgment Accuracy
August 27, 2026
The Think-Probe-Respond method mitigates the tendency of LLMs to default to 'medium novelty' ratings by probing latent judgments from hidden states. This approach improves novelty judgment performance by 22.30% compared to standard baselines.
HOW THIS AFFECTS YOU
●
researcherYou can use this method to create more calibrated automated evaluators for scientific idea generation.