MSAILMichigan Student AI Lab
Reading groups, build-from-scratch projects, and a 37-talk archive, run by students at the University of Michigan since 2008.
The M is a live rectified flow, the sampler inside Stable Diffusion 3 and Flux. Random dots are matched to spots in the M by optimal transport, then each walks its straight line: x_t = (1 − t)·x₀ + t·x₁. The shimmer after it lands is Langevin noise keeping the dots alive.
We strive to spread our passion for AI throughout the University of Michigan student body, regardless of demographic or academic standing.
MSAIL was founded in 2008. During the semester we run reading groups and project teams; since 2020 the speaker series has hosted University of Michigan faculty, PhD researchers, and engineering teams from Bloomberg, ProQuest, and Datature.
Initiatives
All initiatives and the archiveCompetitive Build Initiative
Team up with other MSAIL members for AI competitions and long-term hackathons.
AI Research Project Team
Join a small team of student researchers working directly with a lab PI, contributing to a real research project from planning to completion. Projects run at least a semester, with two teams planned per term; this fall's tentative partner is the AIMS lab in Nuclear Engineering and Radiological Sciences. Python experience required, research experience not.
Model Mining: Alignment & Interp
Hands-on work with language models: extracting the behaviors you want, then using evals and mechanistic interpretability to check alignment and understand what is going on inside. Tentative topics include SFT, DPO, RL, causal steering, evals, and patching. Neural network experience required; LLM or math background helps.
AI in the News
Regular discussions of what AI is doing out in the world, from scientific papers to news headlines: the highlights and the hogwash.
37 recorded talks, 2020 to 2024
- AlphaFold 2 and the Protein Folding Problem
- Concrete Problems in AI Safety
- Scaling Neural Tangent Kernels via Sketching and Random Features
- Faculty Talk: Cognitive Architecture
- The Trend Towards Large Language Models
- Fairness in Machine Learning
Find us
MSAIL is open to every University of Michigan student. These are our channels; Slack is the most active.