AI-Sinkhole: Balancing AI in Education

Large language models are revolutionizing education, offering tools for accessibility and personalized learning. But let's not ignore the elephant in the room: they might be compromising academic integrity. AI-Sinkhole is the latest attempt to strike a balance.

The Problem with LLMs

It seems like every week there's a new paper touting the benefits of large language models (LLMs) in education. But ask who funded the study. The reality is more complex. LLMs like LLama 3 and DeepSeek-R1 have the potential to undermine critical thinking. They're enabling students to bypass deep thought, leading to more cognitive offloading. This isn't just a shift in how we learn. it's a shift in what we value.

Introducing AI-Sinkhole

The AI-Sinkhole framework aims to navigate these murky waters. It's an AI-agent augmented DNS-based framework that discovers and blocks LLM chatbot services during exams. The idea is to protect academic rigor while still harnessing AI's benefits.

AI-Sinkhole uses quantized LLMs for explainable classification and employs Pi-Hole for dynamic DNS blocking. The system achieved impressive results, with cross-lingual performance boasting an F1-score over 0.83. Not bad at all. But the real question is, does it go far enough?

Why It Matters

The benchmark doesn't capture what matters most. We're not just talking about academic performance metrics. This is a story about power, not just performance. Who gets to decide how AI is used in education? And whose benefit are we prioritizing?

AI-Sinkhole has its code available for future research and development, a step toward transparency and collaboration. But let's look closer. Until educators and policymakers address these fundamental issues, no amount of technology will solve the core challenges. Whose data? Whose labor? Whose benefit? It's time we ask these tough questions.