MINT: Evaluating LLMs in Multi-turn Interaction with Tools and Language Feedback

By Xingyao Wang, Zihan Wang, Jiateng Liu, Yangyi Chen, Lifan Yuan, Hao Peng, Heng Ji. International Conference on Learning Representations, 2023

342 citations recorded by the source corpus.

To solve complex tasks, large language models (LLMs) often require multiple rounds of interactions with the user, sometimes assisted by external tools. However, current evaluation protocols often emphasize benchmark performance with single-turn exchanges, neglecting the nuanced interactions among the user, LLMs, and external tools, while also underestimating the importance of natural language feedback from users. These oversights contribute to discrepancies between research benchmark evaluations and real-world use cases. We introduce MINT, a benchmark that evaluates LLMs' ability to solve tasks with multi-turn interactions by (1) using tools and (2) leveraging natural language feedback. To ensure reproducibility, we provide an evaluation framework where LLMs can access tools by executing Python code and receive users' natural language feedback simulated by GPT-4. We repurpose a diverse set of established evaluation datasets focusing on reasoning, coding, and decision-making and carefully curate them into a compact subset for efficient evaluation. Our analysis of 20 open- and closed-source LLMs offers intriguing findings. (a) LLMs generally benefit from tools and language feedback, with performance gains (absolute, same below) of 1-8% for each turn of tool use and 2-17% with natural language feedback. (b) Better single-turn performance does not guarantee better multi-turn performance. (c) Surprisingly, on the LLMs evaluated, supervised instruction-finetuning (SIFT) and reinforcement learning from human feedback (RLHF) generally hurt multi-turn capabilities. We expect MINT can help measure progress and incentivize research in improving LLMs' capabilities in multi-turn interactions, especially for open-source communities where multi-turn human evaluation can be less accessible compared to commercial LLMs with a larger user base.

Subfields: Artificial Intelligence, Information Systems

View Paper: https://www.semanticscholar.org/paper/12b233752c7097ea6525622bed238ae2d2193c5a

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Maosong Sun
2. Graham Neubig
3. Heng Ji
4. Hao Peng
5. Yankai Lin
6. Zhiyuan Liu
7. Yangyi Chen
8. Xingyao Wang

In Information Systems

1. Xingyao Wang
2. Graham Neubig
3. Heng Ji
4. Maosong Sun
5. Hao Peng
6. Yankai Lin
7. Ganqu Cui
8. Uri Alon

Citations in the corpus

• A Survey on Evaluation of Large Language Models (2023, neutral): “CMB [199] / MINT [201] / Dialogue CoT [193] / M3Exam [236] / GAOKAO-Bench [231] / SafetyBench [240]”
• UltraFeedback: Boosting Language Models with High-quality Feedback (2023, neutral): “Compared with two critique models, we find that (the community-trained) Shepherd almost always fails to provide high-quality feedback.”
• Executable Code Actions Elicit Better LLM Agents (2024, methodology): “Both CodeActInstruct and general conversations contribute to agent tasks, while general conversations are essential to maintain performance on general tasks.”
• ULTRAFEEDBACK: Boosting Language Models with Scaled AI Feedback (2023, positive): “We also train a critique model stemming from U LTRA F EEDBACK to boost future research in learning from textual feedback (Wang et al., 2023d).”
• OpenHands: An Open Platform for AI Software Developers as Generalist Agents (2024, positive): “MINT (Wang et al., 2024b) is a benchmark designed to evaluate agents’ ability to solve challenging tasks through multi-turn interactions using tools and natural language feedback simulated by GPT-4.”
• LLMs Get Lost In Multi-Turn Conversation (2025, positive): “Follow-up works expanded on MT-bench, for instance to include longer conversations [37, 18], increase evaluation granularity [2], or to tackle different aspects such as naturalness [72] or tool use [85, 80].”
• Tool learning with large language models: a survey (2024, neutral): “As shown in Figure 1, the past year has witnessed a rapid surge in research efforts on tool learning concurrent with the rise of LLMs. Notably, in practical applications, GPT-4 [5] addresses its knowledge limitations and augments its capabilities by calling on plugins, ultimately integrating the…”