Large Language Models Can Take False First Steps at Inference-time Planning
This provides a theoretical explanation for a subtle inconsistency in LLM behavior, which is incremental but addresses a specific issue in AI planning and inference.
The paper tackles the problem of large language models exhibiting short-sighted and inconsistent planning behavior at inference time, despite having sequence-level planning abilities from training, and finds that this is due to a planning-shift driven by accumulated self-generated context, validated through controlled experiments showing constrained planning and reduced initial bias.
Large language models (LLMs) have been shown to acquire sequence-level planning abilities during training, yet their planning behavior exhibited at inference time often appears short-sighted and inconsistent with these capabilities. We propose a Bayesian account for this gap by grounding planning behavior in the evolving generative context: given the subtle differences between natural language and the language internalized by LLMs, accumulated self-generated context drives a planning-shift during inference and thereby creates the appearance of compromised planning behavior. We further validate the proposed model through two controlled experiments: a random-generation task demonstrating constrained planning under human prompts and increasing planning strength as self-generated context accumulates, and a Gaussian-sampling task showing reduced initial bias when conditioning on self-generated sequences. These findings provide a theoretical explanation along with empirical evidence for characterizing how LLMs plan ahead during inference.