Papers by Ayush Garg
Grounded, or a Good Guesser? A Per-Question Balanced Dataset to Separate Blind from Grounded Models for Embodied Question Answering (2025.acl-short)
Copied to clipboard
Miles Shelton, Nate Wingerd, Kritim K Rijal, Ayush Garg, Adelina Gutic, Brett Barnes, Catherine Finegan-Dollak
| Challenge: | Embodied question answering (EQA) is based on using perception and action in an environment to answer natural language questions. |
| Approach: | They propose a "per-question balanced" EQA dataset that uses two different environments to ground a model's answers in its environment. |
| Outcome: | The proposed model performs better than chance on the PQB-EQA benchmark, showing that it does not require the model to use perception, let alone to act in its environment to find the answer. |
IPO: Your Language Model is Secretly a Preference Classifier (2025.acl-long)
Copied to clipboard
| Challenge: | Reinforcement learning from human feedback (RLHF) is the primary method for aligning large language models with human preferences, but it often incurs significant computational and financial costs due to its reliance on training external reward models or human-labeled preferences. |
| Approach: | They propose an alternative approach that leverages generative LLMs as preference classifiers to reduce the dependence on external reward models or human-labeled preferences. |
| Outcome: | The proposed approach reduces the dependence on external reward models or human-labeled preferences by using generative LLMs as preference classifiers. |
The Inner Monologue of Language Models: When Reasoning Traces Reveal More Than They Hide (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in large language models have enabled them to tackle complex tasks . a fundamental question is: are these models aware of what they "learn" and "think"? |
| Approach: | They define three core competencies: awareness of learned latent policies, generalization of these policies across domains, alignment between internal reasoning traces and final outputs. |
| Outcome: | The results show that RL-trained models exhibit stronger generalizability to novel tasks than SFT models but weak alignment between reasoning traces and final outputs. |