ConfidentLM

Democratizing Education with "Self-Aware" Local AI Access to quality education should not depend on your internet connection. ConfidentLM is a research-backed educational tool designed to bring the power of Large Language Models (LLMs) to communities completely disconnected from the cloud. The Challenge: Silent Errors Local AI models are promising for offline learning, but they suffer from "hallucinations"—confidently stating incorrect facts. In an educational setting, these silent errors are dangerous for students who might unknowingly learn false information. 100% Offline: Powered by a Liquid Foundation Model (LFM2-1.2B) , this app runs locally on your device using 4-bit quantization. No data ever leaves your phone.The Solution: ConfidentLM Developed as a Stanford CS229 Machine Learning project, ConfidentLM introduces a safety layer for on-device models. Hallucination Detection: Unlike standard chatbots, ConfidentLM analyzes the model's internal "brain activity" (hidden states) to predict if an answer is likely a hallucination. Trustworthy Tutoring: By flagging unreliable answers, we create a "tutor in your pocket" that is safe enough for real-world learning. Join us in bridging the digital divide with AI that is accessible, efficient, and—most importantly—confident in its accuracy.