A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity
By Yejin Bang, Samuel Cahyawijaya, Nayeon Lee, Wenliang Dai, Dan Su, Bryan Wilie, Holy Lovenia, Ziwei Ji, Tiezheng Yu, Willy Chung and 3 more. International Joint Conference on Natural Language Processing, 2023
1,776 citations recorded by the source corpus.
This paper proposes a framework for quantitatively evaluating interactive LLMs such as ChatGPT using publicly available data sets. We carry out an extensive technical evaluation of ChatGPT using 23 data sets covering 8 different common NLP application tasks. We evaluate the multitask, multilingual and multi-modal aspects of ChatGPT based on these data sets and a newly designed multimodal dataset. We find that ChatGPT outperforms LLMs with zero-shot learning on most tasks and even outperforms fine-tuned models on some tasks. We find that it is better at understanding non-Latin script languages than generating them. It is able to generate multimodal content from textual prompts, via an intermediate code generation step. Moreover, we find that ChatGPT is 63.41% accurate on average in 10 different reasoning categories under logical reasoning, non-textual reasoning, and commonsense reasoning, hence making it an unreliable reasoner. It is, for example, better at deductive than inductive reasoning. ChatGPT suffers from hallucination problems like other LLMs and it generates more extrinsic hallucinations from its parametric memory as it does not have access to an external knowledge base. Finally, the interactive feature of ChatGPT enables human collaboration with the underlying LLM to improve its performance, i.e, 8% ROUGE-1 on summarization and 2% ChrF++ on machine translation, in a multi-turn"prompt engineering"fashion. We also release codebase for evaluation set extraction.
Subfields: Applied Psychology, Health Informatics, Social Psychology
View Paper: https://www.semanticscholar.org/paper/bf8491bef353df126e2306ad2fe4b898697b906a
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Applied Psychology
In Health Informatics
- Heng Ji
- Matei A. Zaharia
- Alexander J. Ratner
- Siyuan Ma
- James Y. Zou
- Zheng-Xin Yong
- Mihai Surdeanu
- Nikolaos D. Tselikas
In Social Psychology
Citations in the corpus
- LLM Lies: Hallucinations are not Bugs, but Features as Adversarial Examples (2023, background): “However, we still can not completely trust their answers, LLMs suffer from hallucinations [2, 8] which means LLMs lie and fabricate non-existent facts or inappropriate information.”
- Low-Resource Languages Jailbreak GPT-4 (2023, result): “More importantly, our results also suggest that researchers have underestimated the capability of the LLMs in understanding and generating text in low-resource languages [4, 18, 22, 39, 46, 48].”
- Fuzz4ALL: Universal Fuzzing with Large Language Models (2023, methodology): “More recently, instruction-based LLMs (e.g., ChatGPT [65] and GPT4 [55]) and LLMs fine-tuned using re-inforcement learning from human feedback (RLHF) [88] are shown to understand and follow complex instructions [4, 56, 65].”
- EvalCrafter: Benchmarking and Evaluating Large Video Generation Models (2023, background): “Similarly, the evaluation of the multi-modal generative model also draws the attention of the researchers [7,62].”
- Evaluating Large Language Models in Class-Level Code Generation (2024, background): “To date, the evaluation for LLMs has covered a wide range [20], encompassing not only traditional NLP tasks ( e.g., sentiment analysis [17], question answering [16], and reasoning [19]) but also some specific downstream domains ( e.g., medicine [23], agent [34] and recommendation system [28]).”
- Time Travel in LLMs: Tracing Data Contamination in Large Language Models (2023, background): “This shift has led to an extensive range of LLMs (Touvron et al. 2023a;b; Biderman et al. 2023; K¨opf et al. 2023; Chung et al. 2022; Penedo et al. 2023, inter-alia) which excel in various professional and academic benchmarks (Bang et al. 2023; Bubeck et al. 2023).”
- HaluEval: A Large-Scale Hallucination Evaluation Benchmark for Large Language Models (2023, background): “2023; Bang et al., 2023; Sun et al., 2023), where the generated content is either in conflict with existing source or cannot be verified by the available knowledge resources.”
- Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models (2023, neutral): “…for a single task, and thus, hallucination studies on them are usually task-specific (Maynez et al., 2020; Wang and Sennrich, 2020; Xiao and Wang, 2021); however, current LLMs are expected to excel in multi-task, multi-lingual, and multi-domain settings (Bang et al., 2023; Chang et al., 2023).”
- The rise and potential of large language model based agents: a survey (2025, neutral): “Recently, the rise and development of Transformer-based generative LLMs have endowed LLM-based agents with inherent language generation capabilities [19,224].”
- ChatGPT Beyond English: Towards a Comprehensive Evaluation of Large Language Models in Multilingual Learning (2023, methodology): “Beyond English, the analysis of ChatGPT’s responses for input questions in (Guo et al., 2023) is only done for Chinese, while the results of the medical licensing examinations for ChatGPT are only shown for Japanese in (Kasai et al., 2023).”
- Unifying Large Language Models and Knowledge Graphs: A Roadmap (2023, background): “However, further studies reveal that LLMs are not able to recall facts and often experience hallucinations by generating statements that are factually incorrect [15], [28].”
- Evaluating Object Hallucination in Large Vision-Language Models (2023, background): “Especially, LLMs tend to hallucinate unintended text (Huang et al., 2021; Bang et al., 2023), and VLPMs might generate nonexistent objects in the image (Biten et al., 2022) (termed as object hal-lucination ).”
- An Embodied Generalist Agent in 3D World (2023, neutral): “Answer Refinement There are 3 curtains in the room.”
- Is ChatGPT a Good NLG Evaluator? A Preliminary Study (2023, background): “…obtained promising results on various natural language understanding (NLU) and natural language generation (NLG) downstream tasks (Ouyang et al., 2022; Koco ´ n et al., 2023; Qin et al., 2023; Huang et al., 2023; Yang et al., 2023; Rao et al., 2023; Bang et al., 2023; Zuccon and Koopman, 2023).”
- Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate (2023, background): “…like Chat-GPT, GPT-4 (OpenAI, 2023) and Bard 1 , have shown remarkable performance on general language tasks (Jiao et al., 2023; Wu et al., 2023; Bang et al., 2023) but still struggle on complex reasoning tasks (Zhu et al., 2022; Gou et al., 2023), which drives the research on cognitive…”
- A Survey of Large Language Models (2023, neutral): “For example, LLMs exhibit a tendency to generate hallucinations [344], which are texts that seem plausible but may be factually incorrect.”