BDH-CQ: In-Context Learning with Recurrent Latent Reasoning
A 150M-parameter BDH-CQ model reports 29.5% pass@2 on ARC-AGI-1 at $0.0007 per task.
The paper says BDH-CQ updates recurrent memory from inference-time examples, then solves queries through latent iterative computation without spelling out intermediate reasoning. The authors test it on ARC-AGI-1 and on controlled ARC-like interventions meant to probe what the model learns from demonstrations. They claim this result breaks the prior ARC-AGI-1 cost-accuracy Pareto frontier and sets a new benchmark cost-efficiency mark. Source: HF Daily Papers' note.
The paper says BDH-CQ updates recurrent memory from inference-time examples, then solves queries through latent iterative computation without spelling out intermediate reasoning. The authors test it on ARC-AGI-1 and on controlled ARC-like interventions meant to probe what the model learns from demonstrations. They claim this result breaks the prior ARC-AGI-1 cost-accuracy Pareto frontier and sets a new benchmark cost-efficiency mark. Source: HF Daily Papers' note.
score 6