NATURAL PLAN: Benchmarking LLMs on Natural Language Planning

TL;DR

NATURAL PLAN is a new benchmark for evaluating large language models' natural language planning abilities across tasks like trip and meeting planning, revealing significant challenges and performance gaps in current models.

Contribution

The paper introduces NATURAL PLAN, a realistic natural language planning benchmark with multiple tasks and contexts, and provides comprehensive evaluation of state-of-the-art LLMs on these tasks.

Findings

Models perform poorly on complex planning tasks, with success rates below 5% for larger problems.

Performance drops significantly as task complexity increases.

Self-correction and in-context learning methods have limited impact on improving planning accuracy.

Abstract

We introduce NATURAL PLAN, a realistic planning benchmark in natural language containing 3 key tasks: Trip Planning, Meeting Planning, and Calendar Scheduling. We focus our evaluation on the planning capabilities of LLMs with full information on the task, by providing outputs from tools such as Google Flights, Google Maps, and Google Calendar as contexts to the models. This eliminates the need for a tool-use environment for evaluating LLMs on Planning. We observe that NATURAL PLAN is a challenging benchmark for state of the art models. For example, in Trip Planning, GPT-4 and Gemini 1.5 Pro could only achieve 31.1% and 34.8% solve rate respectively. We find that model performance drops drastically as the complexity of the problem increases: all models perform below 5% when there are 10 cities, highlighting a significant gap in planning in natural language for SoTA LLMs. We also conduct extensive ablation studies on NATURAL PLAN to further shed light on the (in)effectiveness of approaches such as self-correction, few-shot generalization, and in-context planning with long-contexts on improving LLM planning.