LEAD: Breaking the No-Recovery Bottleneck in Long-Horizon Reasoning
This addresses a critical problem in AI for improving reliability in complex reasoning tasks, though it is incremental as it builds on decomposition methods.
The paper tackles the instability in long-horizon reasoning for Large Language Models by identifying a 'no-recovery bottleneck' caused by non-uniform error distribution, and proposes LEAD to enable the o4-mini model to solve Checkers Jumping up to complexity n=13, improving from failure beyond n=11.
Long-horizon execution in Large Language Models (LLMs) remains unstable even when high-level strategies are provided. Evaluating on controlled algorithmic puzzles, we demonstrate that while decomposition is essential for stability, extreme decomposition creates a "no-recovery bottleneck". We show that this bottleneck becomes critical due to highly non-uniform error distribution, where consistent errors on a few "hard" steps become irreversible. To address this, we propose Lookahead-Enhanced Atomic Decomposition (LEAD). By incorporating short-horizon future validation and aggregating overlapping rollouts, LEAD provides enough isolation to maintain stability while retaining enough local context to correct errors. This enables the o4-mini model to solve Checkers Jumping up to complexity $n=13$, whereas extreme decomposition fails beyond $n=11$.