Papers by Mo Vazifeh
seqBench: A Tunable Benchmark to Quantify Sequential Reasoning Limits of LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | **seqBench** allows systematic variation of several key complexity dimensions. |
| Approach: | They introduce a parametrized benchmark for probing sequential reasoning limits in Large Language Models through precise, multi-dimensional control over several key complexity dimensions. |
| Outcome: | The framework allows systematic variation of logical depth, backtracking requirements and noise ratio on state-of-the-art LLMs. |