πArXiv AIβ’Stalecollected in 17h
LaTA: FERPA-Compliant Local LLM Autograder
π‘Open-source local LLM autograder grades 200 STEM students with <0.05% errorsβFERPA safe!
β‘ 30-Second TL;DR
What Changed
Runs entirely on local hardware like Mac Studio at $0 marginal cost
Why It Matters
LaTA enables privacy-focused AI grading in academia, slashing TA workload and enabling regrading, while demonstrating superior student outcomes.
What To Do Next
Clone the AGPLv3 LaTA repo from arXiv links and test on your LaTeX STEM assignments.
Who should care:Researchers & Academics
Key Points
- β’Runs entirely on local hardware like Mac Studio at $0 marginal cost
- β’Four-stage pipeline: ingest, segment, grade, report using gpt-oss:120b LLM
- β’Binary scoring via YAML rubric comparing to instructor reference solution
- β’Achieved 0.02-0.04% error rate; improved student performance vs prior cohort
π°
Weekly AI Recap
Read this week's curated digest of top AI events β
πRelated Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: ArXiv AI β