XAI meets LLMs: A Survey of the Relation between Explainable AI and Large Language Models

Interpretability is as important as performance: The survey emphasizes that advancing LLM capabilities should not come at the expense of understanding how these models work. Current research often prioritizes functional improvements, but the authors argue for a balanced approach where transparency and explainability receive equal weight to performance gains.

Transparency is increasingly demanded by stakeholders: Both the AI research community and business sectors are recognizing the critical need for transparency in LLMs. This growing interest signals that explainability is no longer optional but essential for responsible deployment of these powerful systems.

XAI and LLM research should advance together: Rather than treating explainability as an afterthought, the survey advocates for integrating XAI methods directly into LLM development. The authors conclude by urging the research community to develop these fields in tandem rather than separately.

Current LLM research follows two divergent paths: The survey identifies a split in how researchers approach LLMs—some focus on enhancing performance while others concentrate on model interpretability. This division suggests the field lacks unified standards for what responsible LLM development should prioritize.

Rapid development creates documentation challenges: By including both peer-reviewed and preprint papers, the survey acknowledges that LLM research is evolving so quickly that traditional peer review alone cannot capture the full landscape, highlighting the need for ongoing monitoring of responsible AI practices.