LLM+XAI Reading Group

In this reading group, each person presents a paper on a topic in LLMs or XAI that he or she wanted to learn any way (since teaching is one of the best ways to learn the topic). Preparing slides is not mandatory; one can just share the paper on a screen. The meetings will take place on Zoom.

I limited the scope of the topic to LLMs and XAI, because

1. They are (still) important (and 🔥)
2. Limiting the scope allows us to focus and perhaps even collaborate on a paper.
3. I like these topics

Potential topics would be:

LLM-related: LLM reasoning, LLMs and causality, LLM evaluation, LLM training, RLHF, RLAIF, applications of LLMs in robotics and other areas, explaining LLMs, LLMs as agents, new LLM architectures (e.g., state-space models), vision-language models (VLMs) …

XAI-related: Novel XAI methods (e.g., mechanistic interpretability), XAI evaluation, applications of XAI, explainable RL, LLMs for XAI, human-centered XAI, …

Upcoming Presentations

Time: Every Wednesday 2pm,

2024

• 25.09 A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicity to be presented by Xintong

Past Presentations

2024

• 18.09 No meeting
• 11.09 Understanding Social Reasoning in Language Models with Language Models presented by Burak
• 04.09 Do Large Language Models Latently Perform Multi-Hop Reasoning? presented by Jae
• 21.08 A Multimodal Automated Interpretability Agent presented by Wenhao
• 14.08 A Concept-Based Explainability Framework for Large Multimodal Models presented by Jae
• 07.08 Advancing LLM Reasoning Generalists with Preference Trees presented by Hassan
• 31.07 STaR: Self-Taught Reasoner presented by Imran
• 24.07 Neuron to Graph: Interpreting Language Model Neurons at Scale presented by Sergio
• 17.07: Detecting hallucinations in large language models using semantic entropy to be presented by Xiaowen
• 10.07: Identifying Linear Relational Concepts in Large Language Models presented by Jae
• 26.06: What does the Knowledge Neuron Thesis Have to do with Knowledge presented by Xintong
• 19.06: LLMs with Chain-of-Thought Are Non-Causal Reasoners presented by Wenhao
• 12.06: Self-Supervised Alignment with Mutual Information: Learning to Follow Principles without Preference Labels presented by Xufeng
• 05.06: A Survey on Evaluation of Large Language Models presented by Hassan
• 29.05: Inference-Time Intervention: Eliciting Truthful Answers from a Language Model presented by Sergio
• 22.05: Finding Neurons in a Haystack: Case Studies with Sparse Probing presented by Imran
• 15.05: Towards Uncovering How Large Language Model Works: An Explainability Perspective presented by Jae