10 adversarial scenarios, 64 assertions, 3-tier evaluation pyramid. Llama, Qwen, GPT-OSS — none scored above 63%. Here's what broke them.
I Built an Adversarial Eval Framework and Attacked 5 LLMs — Every Single One Failed
Related
Self-healing CI vs AI code fixes: which failures belong where
Self-healing CI vs AI code fixes is a question about failure class, not capability: which failures a runner absorbs, and which belong to the agent.
DoseVoice: I built an AI that reads pill bottles out loud (and knows when to say 'turn the bottle')
This is a submission for the MLH x DEV Writing Challenge Built in 36 hours at HackUMass XIV. Try it:...
Ghost Rovers: I Called 12 Machines NASA Left Behind. One Picked Up.
A 1969 Mission Control console for kids. Built in one day for NASA Space Apps 2026. TL;DR: For NASA...