Update README.md

README.md

@@ -30,14 +30,26 @@ configs:
 ---
 **Super easy task for humans**  that **All SOTA LLM fail** to retrieve the correct answer from context. Including SOTA models: GPT5, Grok4, DeepSeek, Gemini 2.5PRO, Mistral, Llama4...etc 
 
+- **Update: Accepted to COLM 2026 (San Francisco).**
 - **AAAI 2026 Worshop Oral**: Jan/2026 LaMAS (LLM-based Multi-Agent Systems: Towards Responsible, Reliable, and Scalable Agentic Systems) Jan/2026 Singapole
 - ICML 2025 Long-Context Foundation Models Workshop Accepted.(https://arxiv.org/abs/2506.08184)
+
 - Update: This dataset is integrated into Moonshot AI(Kimi)'s **internal benchmarking framework** for assessing ** tracking capacity and context interference in LLM/agents**.
 > **Update1:mergerd into Moonshot/Kimi AI's internal eval tools and under review by a xAI(Grok)'s' eval team**
 > **Update2:AAAI 2026 workshop oral Jan/2026. Singapole for the updated paper version**
 > 
 
 
+
+## Contact & Collaboration
+
+I welcome discussions and collaborations with research labs and model teams on **mechanistic interpretability—especially how LLMs represent, update, and lose information—rigorous model and agent evaluation, and why models fail, not only whether they fail**. If you are interested in PI-LLM, benchmark comparisons, or evaluating your models or agents, please reach out.
+
+**Email:** [wangchupei@hotmail.com](mailto:wangchupei@hotmail.com)  
+
+I usually reply within 36 hours.
+
+
 ## Key–value update paradigm (what the model sees) 1 keys, N updates each
 
 ```