# Hypothesis-only Biases in Large Language Model-Elicited Natural Language Inference

arXiv: https://arxiv.org/abs/2410.08996 · HTML: https://mlsanity.com/paper/2410.08996/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 25 of its 42 references that are in the map.
- First posted on arXiv 11 Oct 2024.

## Builds on (most cited first, 25 of 25)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Self-Instruct: Aligning Language Models with Self-Generated Instructions](https://mlsanity.com/paper/2212.10560/) (2022)
- [BLOOM: A 176B-Parameter Open-Access Multilingual Language Model](https://mlsanity.com/paper/2211.05100/) (2022)
- [A large annotated corpus for learning natural language inference](https://mlsanity.com/paper/1508.05326/) (2015)
- [SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems](https://mlsanity.com/paper/1905.00537/) (2019)
- [In-context Learning and Induction Heads](https://mlsanity.com/paper/2209.11895/) (2022)
- [ChatGPT Outperforms Crowd-Workers for Text-Annotation Tasks](https://mlsanity.com/paper/2303.15056/) (2023)
- [Annotation Artifacts in Natural Language Inference Data](https://mlsanity.com/paper/1803.02324/) (2018)
- [OLMo: Accelerating the Science of Language Models](https://mlsanity.com/paper/2402.00838/) (2024)
- [The Falcon Series of Open Language Models](https://mlsanity.com/paper/2311.16867/) (2023)
- [Benchmarking Large Language Models for News Summarization](https://mlsanity.com/paper/2301.13848/) (2023)
- [Unnatural Instructions: Tuning Language Models with (Almost) No Human Labor](https://mlsanity.com/paper/2212.09689/) (2022)
- [Hypothesis Only Baselines in Natural Language Inference](https://mlsanity.com/paper/1805.01042/) (2018)
- [Polyjuice: Generating Counterfactuals for Explaining, Evaluating, and Improving Models](https://mlsanity.com/paper/2101.00288/) (2021)
- [AnnoLLM: Making Large Language Models to Be Better Crowdsourced Annotators](https://mlsanity.com/paper/2303.16854/) (2023)
- [Reframing Human-AI Collaboration for Generating Free-Text Explanations](https://mlsanity.com/paper/2112.08674/) (2021)
- [Increasing Diversity While Maintaining Accuracy: Text Data Generation with Large Language Models and Human Interventions](https://mlsanity.com/paper/2306.04140/) (2023)
- [Artificial Artificial Artificial Intelligence: Crowd Workers Widely Use Large Language Models for Text Production Tasks](https://mlsanity.com/paper/2306.07899/) (2023)
- [How Much Reading Does Reading Comprehension Require? A Critical Investigation of Popular Benchmarks](https://mlsanity.com/paper/1808.04926/) (2018)
- [Rethinking the Role of Scale for In-Context Learning: An Interpretability-based Case Study at 66 Billion Scale](https://mlsanity.com/paper/2212.09095/) (2022)
- [Performance Impact Caused by Hidden Bias of Training Data for Recognizing Textual Entailment](https://mlsanity.com/paper/1804.08117/) (2018)
- [The Effect of Different Writing Tasks on Linguistic Style: A Case Study of the ROC Story Cloze Task](https://mlsanity.com/paper/1702.01841/) (2017)
- [Large Language Models as Annotators: Enhancing Generalization of NLP Models at Minimal Cost](https://mlsanity.com/paper/2306.15766/) (2023)
- [Genie: Achieving Human Parity in Content-Grounded Datasets Generation](https://mlsanity.com/paper/2401.14367/) (2024)

## Abstract (arXiv)

We test whether replacing crowdsource workers with LLMs to write Natural Language Inference (NLI) hypotheses similarly results in annotation artifacts. We recreate a portion of the Stanford NLI corpus using GPT-4, Llama-2 and Mistral 7b, and train hypothesis-only classifiers to determine whether LLM-elicited hypotheses contain annotation artifacts. On our LLM-elicited NLI datasets, BERT-based hypothesis-only classifiers achieve between 86-96% accuracy, indicating these datasets contain hypothesis-only artifacts. We also find frequent "give-aways" in LLM-generated hypotheses, e.g. the phrase "swimming in a pool" appears in more than 10,000 contradictions generated by GPT-4. Our analysis provides empirical evidence that well-attested biases in NLI can persist in LLM-generated data.