ResearchNovember 12, 2026
-45
🧊
Two LLMs Now Grade Orally
A research paper introduces a framework that simulates Islamic jurisprudential rulings using finite state automata and large language models.
#LLM#education#multi-agent#assessment#arXiv

🔥 What happened
University of Illinois researchers unveiled a chat-based exam system that mimics oral exams using two LLM agents: an interviewer asks questions while a hidden evaluator tracks a rubric and only nudges the interviewer. They tested it against simulated student personas — including prompt-injectors and answer-extractors — and deployed it in a real AI literacy course.
💡 Why it matters
Oral exams are pedagogically gold but don't scale — a human examiner handles maybe 20 a day. The two-agent split keeps answers from leaking to the interviewer and could assess thousands of students at once. The catch: the authors themselves found gaps in follow-up questioning and grading credit.
⚡ Our take
The architecture is the real innovation here, not the model. Anyone automating assessment without separating dialogue from evaluation is building a leak machine.
The title, summary and analysis of this item were produced automatically by an AI system and have not been editorially reviewed. They may contain errors, bias or omissions — when in doubt, read the linked original source.
Deep Dives & Similar Intelligence

Rewriting boosts LLMs by 27 percent
ResearchNovember 11, 2026

PreG Slashes Reasoning Token Costs
ResearchNovember 11, 2026
Alibaba open-sources its code review agent
Tools & ProjectsNovember 13, 2026
Context Windows Obsolete: Billion-Token Sessions
Tools & ProjectsNovember 11, 2026

MCP Burns 59K Tokens Per Task
Tools & ProjectsNovember 11, 2026