Research, Interpretability, Code
March 2026
Your LLM Understood the Code — Then Forgot the Answer
We use Patchscopes to dissect how Code LLMs process code layer-by-layer, revealing a counterintuitive phenomenon: models often compute the right answer mid-forward-pass, then actively overwrite...
Read More
Research
March 2025
NeurIPS 2025 SE-Agent Case Study
A detailed analysis and case study of the SE-Agent performance at NeurIPS 2025, exploring multimodal learning and agentic workflows.
Read More
Opinion
Upcoming
Future Research Directions in Multimodal Agents
Coming soon... I'll be sharing some insights on the next generation of
multimodal agents and their applications.
Stay tuned