The CoTu team developed a neuro-symbolic Program-of-Thought pipeline for the EXACT 2026 competition, which requires transparent educational question answering using small, self-hosted language models. Their system, based on a 4B parameter model, generates Z3 encodings for logical reasoning and Python code for physics problems, both wrapped in a self-correction loop and explained-JSON output. This approach achieved a perfect score on the physics task and the highest technical score in the final round, demonstrating that grounding answers in symbolic solvers enables correct deductions even with smaller models. AI
IMPACT Demonstrates effective neuro-symbolic reasoning with small models for transparent educational QA, potentially influencing future research in explainable AI.
RANK_REASON The cluster describes a research paper detailing a system developed for a specific competition, including its methodology and results.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
