How to Answer Reviewers Who Say Your Limitations Discussion Is Superficial

Why superficial limitation discussions are problematic

Reviewers often criticize papers for stating limitations only in generic terms such as insufficient compute or time constraints, indicating that the authors may not fully understand the method’s boundaries or real‑world applicability.

❌ Common weak reply

We acknowledge the limitation that our method requires considerable computational resources and runtime, and we will optimize the structure in future work.

This answer is rejected because it admits a limitation without providing new information or concrete details.

What reviewers actually want

Where exactly do the limitations appear?

Are they due to model capacity, data dependence, model size, or deployment cost?

In which scenarios do these limitations affect performance?

Could they cause negative transfer, robustness loss, or a ceiling effect?

Is there a clear direction for future improvement?

✅ Recommended reply structure

Thank you for the insightful comment; we agree that the discussion of limitations in the current version is insufficient. In the revised manuscript we will add a “Limitations and Future Work” section covering the following points: 1. Our approach relies on the base model’s initial capability. When the base model is weak, adding complex modules may interfere with its representations and even cause negative transfer. 2. The non‑parametric extraction is still bounded by the model’s cognitive ceiling. Future work will explore reinforcement‑learning paradigms to push this boundary. 3. In real applications, the method may be impacted by computational overhead, domain shift, and deployment efficiency; we plan to further optimize these aspects.

🎯 Core answer strategy

Instead of a generic statement, explicitly detail the limitation by addressing three aspects:

1. Explain the ability boundary

The method depends on the base model’s existing abilities; if the base model is under‑performing, the added modules may provide limited gains.

2. Describe potential risks

Overly complex additional modules can interfere with the model’s original representations, leading to negative transfer.

3. Outline future directions

Future work may combine reinforcement learning, parameter‑efficient adaptation, or lighter architectures to mitigate current limitations.

Following this structured, specific approach demonstrates a deep understanding of the method’s boundaries and real‑world implications, satisfying reviewer expectations.