Show HN: Training a SOTA Pangram-like detector on a 4070ti with Claude

Category: ai-ml

Tags: ai-detection, fine-tuning, llm, text-classification, model-card, autonomous-agents

Score: 6.7/10 (Innovation: 6, Technical: 8, Documentation: 7, Utility: 6)

A fine-tuned 4B-parameter open-weight AI text detector that classifies English text as human-written, AI-written, or AI-edited on a per-sentence basis, benchmarked competitively against Pangram. Notably the entire project—data collection, training, evaluation, and model card—was autonomously generated by Claude Opus, making it an interesting demonstration of agent-driven ML research.

Target audience: researchers, educators, ML engineers, content moderators

Repository: https://huggingface.co/mesmertech/mesmer-ai-detector-4b

View on Hacker News