Papers by Kv Aditya Srivatsa
LLMs cannot spot math errors, even when allowed to peek into the solution (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) demonstrate impressive performance on existing reasoning benchmarks, but struggle with meta-reasoning tasks such as locating the first error step in student solutions. |
| Approach: | They propose an approach that generates an intermediate corrected student solution, aligning more closely with the original student’s solution, which helps improve performance. |
| Outcome: | The proposed approach generates an intermediate corrected student solution, aligning more closely with the original student’s solution, which helps improve performance. |
SelectLLM: Query-Aware Efficient Selection Algorithm for Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing large language models struggle with complex tasks such as factually-grounded reasoning and planning due to inherent training biases, model size constraints, and the quality or diversity of pre-training datasets. |
| Approach: | They propose a novel algorithm to select the most suitable LLMs from a large pool and use it to efficiently generalize and perform tasks. |
| Outcome: | The proposed model outperforms existing ensemble-based baselines and achieves competitive performance with similarly sized top-performing LLMs while maintaining efficiency. |
What Makes Math Word Problems Challenging for LLMs? (2024.findings-naacl)
Copied to clipboard
| Challenge: | Experiments show that even quite powerful LLMs are still challenged by MWPs. |
| Approach: | They propose to analyze what makes math word problems (MWPs) in English challenging for large language models (LLMs). |
| Outcome: | The proposed model can handle a range of core NLP tasks, but it has emergent abilities, such as ability to solve mathematical puzzles. |
Unifying AI Tutor Evaluation: An Evaluation Taxonomy for Pedagogical Ability Assessment of LLM-Powered AI Tutors (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing evaluations of large language models have been limited to subjective protocols and benchmarks. |
| Approach: | They propose a unified evaluation taxonomy with eight pedagogical dimensions based on key learning sciences principles to assess the pedagical value of LLM-powered AI tutor responses grounded in student mistakes or confusions in the mathematical domain. |
| Outcome: | The proposed taxonomy, benchmark, and human-annotated labels will streamline the evaluation process and help track the progress in AI tutors’ development. |