Notes on putting AI to work in the enterprise: responsibly, and without slowing down.
Recent writing
- GLM 5.3, Open Weights, and Chaining Vulnerabilities
What Anthropic's GLM 5.3 findings mean when an advanced model that can chain vulnerabilities ships with open weights.
- Testing Jev on Prompt Injection Detection
A weekend test of Jev against seven other models across four public prompt-injection datasets.
- TypeSafe AI's Jev: A Model That Outputs Numbers, Not Text
Jev skips the text-to-structure translation step, which could mean faster, cheaper, and more auditable AI decisions.
Projects
- Read MoreSocial book-tracking app for tracking reading and organizing shelves.
- journl/itZero-knowledge encrypted journaling app.
- How Do GenAI Models Evaluate Other Models?Using generative AI models to grade other models' outputs.
- Using BigQuery as a Vector StoreTutorial on using BigQuery as a vector store for RAG applications.
- Build a Card Game With GenAIBuilding and deploying a War card game with Bolt.new in minutes.
- Simplify Testing Generative AI SystemsA simple way to test the subjective outputs of generative AI systems.
- Quick QuestionRandom questions to ask friends and family.