Summary of Llms Still Can’t Plan; Can Lrms? a Preliminary Evaluation Of Openai’s O1 on Planbench, by Karthik Valmeekam et al.
LLMs Still Can’t Plan; Can LRMs? A Preliminary Evaluation of OpenAI’s o1 on PlanBench
by Karthik Valmeekam, Kaya Stechly, Subbarao Kambhampati
First submitted to arxiv on: 20 Sep 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Computation and Language (cs.CL)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper explores the planning abilities of large language models (LLMs) and their Large Reasoning Models (LRMs). It uses PlanBench, a benchmark developed in 2022 to evaluate LLMs’ planning capabilities. The authors investigate the performance of current LLMs and new LRMs on this benchmark, finding that while the o1 model shows significant improvement, it still falls short of saturating the benchmark. The paper raises questions about accuracy, efficiency, and guarantees, emphasizing the importance of considering these factors before deploying such systems. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how well big language models can plan for things to happen. It uses a special tool called PlanBench to test these models. The authors are curious about whether new types of models, like Large Reasoning Models (LRMs), can do better on this benchmark than the current ones. They find that one new model, o1, does really well, but there’s still room for improvement. This makes them wonder if we should worry about things like getting the right answer and using the right amount of energy when we use these models. |
