Papers by Gauri Kambhatla
Promoting Constructive Deliberation: Reframing for Receptiveness (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Current methods for promoting pro-social discussion and debate online are limited. |
| Approach: | They propose automatic reframing of disagreeing responses to signal receptiveness to a preceding comment. |
| Outcome: | The proposed framework can be used to promote constructive debate and debate online. |
Measuring Lexical Diversity of Synthetic Data Generated through Fine-Grained Persona Prompting (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Fine-grained personas have been used for generating ‘diverse’ synthetic data for pre-training and supervised fine-tuning of Large Language Models (LLMs). |
| Approach: | They measure the diversity of persona-driven synthetically generated prompts and responses with a suite of lexical diversity and redundancy metrics. |
| Outcome: | The proposed model is based on human-written prompts and responses, but human-generated prompts are significantly less diverse than human-created ones. |
Quantifying Train-Evaluation Overlap with Nearest Neighbors (2023.findings-acl)
Copied to clipboard
| Challenge: | Using a novel metric, we quantify train-evaluation overlap in benchmark datasets to evaluate model generalization. |
| Approach: | They quantify train-evaluation overlap as a measure of an individual dataset’s adequacy to evaluate model generalization over a wide range of datasets. |
| Outcome: | The proposed metric quantifies train-evaluation overlap, providing insights for constructing datasets to study generalization. |
Improving the Distributional Alignment of LLMs using Supervision (2026.acl-long)
Copied to clipboard
Gauri Kambhatla, Sanjana Gautam, Angela Zhang, Alexander Liu, Ravi Srinivasan, Junyi Jessy Li, Matthew Lease
| Challenge: | Existing work to evaluate LLMs' alignment with human values and opinions has a key shortcoming. |
| Approach: | They propose to add supervision to LLMs to improve alignment with diverse populations . they find that supervision improves alignment across public health, public opinion, values and beliefs . |
| Outcome: | The proposed method improves the alignment of LLMs with diverse populations on subjective questions. |