Question: is WFGY a suitable practical resource for the XAI list #39
Description
Hi,
Thank you for maintaining Awesome-explainable-AI. This is one of the few places where the recent survey papers and key references on XAI and interpretability are collected in a structured way. I have used it several times to explore the LLM and NLP survey space.
I maintain an open source project that is not a research paper, but might be relevant as a practical resource for studying failure and interpretability of LLMs in extreme conditions. I wanted to ask whether it is in scope before proposing any change.
Project: WFGY
Repo: https://github.com/onestardao/WFGY
Very short description:
WFGY is an open source framework for debugging and stress testing large language models under long horizon, high tension text scenarios. Instead of focusing on single predictions, it exposes how a model behaves over a long sequence of tightly coupled prompts.
WFGY 3.0 Singularity Demo packages a family of one hundred plus stress-test questions into a single TXT file. When you upload it to a model and follow the menu, it produces traces that make it easy to see where the model’s reasoning drifts, collapses, or recovers. This has been useful for practitioners who want to observe model behavior in a more interpretable way, without changing the model weights.
I realize that this repository is mainly focused on papers and formal surveys. If you feel that practical tools like WFGY do not belong here, I fully understand. If you think it could be listed somewhere as an applied resource for LLM behavior and explanation studies, I would be happy to suggest a minimal addition, for example in an “others” or “applied resources” section if you plan to keep such a category.
Either way, thank you for curating this list. It has already been useful for my own reading.