Articles
All published articles, newest first.
Testing AI
Metric of the Week: G-Eval
How G-Eval turns an LLM into a scorer for open-ended text: the form-filling paradigm, probability-weighted scores, and what the SummEval numbers show.
Sun Sep 13 2026
Testing AI
LLM-as-a-Judge: All Eyez on... LLM
The judge has priors: twelve documented biases in LLM-as-a-judge, what the research actually shows, and where a real fix exists.
Sun Sep 06 2026
AI Security
PGD: Same Attack, Way More Persistent
How Projected Gradient Descent iterates past FGSM's single-step limits, with real captured results including a targeted attack that actually lands.
Wed Aug 19 2026
AI Security
FGSM: Story of how I tricked a neural net with one gradient step
What the Fast Gradient Sign Method is, how it works, and how to run it yourself against a real ImageNet classifier, with code.
Sat Aug 15 2026