Does an LLM Watermark Survive Editing? Testing 4 Schemes Against ParaphrasingWe test a green-red list LLM watermark against word edits, synonym swaps and LLM paraphrasing and compare four seeding schemes: robustness vs security24 min read·milosbozic.hashnode.devSep 14
Building a Reliable AI Agent from Scratch with a Small Local LLM (Part 2)Tools, parallel calls and LangGraph: from 5 to 12 correct answers19 min read·milosbozic.hashnode.devSep 10
How LLM Text Watermarking Works: Build One from Scratch in PythonBuild the green-red list watermark for LLM text in Python with Hugging Face, detect it with a z-test, and measure its cost in text quality.31 min read·milosbozic.hashnode.devSep 8
Building a Reliable AI Agent from Scratch with a Small Local LLMPrompts, reflection and constrained decoding, each measured on the same test set22 min read·milosbozic.hashnode.devSep 3