The center of gravity in AI security shifted from making models say bad text to making systems do bad actions. Here's what changed, why it matters, and what defenses actually work when agents can call tools, delegate across workers, and act on enterprise systems.
Continue ReadingA position paper arguing that predictions used by learning-augmented algorithms should be treated as untrusted algorithmic inputs, not neutral estimates. Maps adversarial machine learning threat models into algorithm design and proposes benchmark dimensions for robust, security-aware online decision-making.
Continue ReadingHouse party for Claude Code enjoyers. Salespeople who now ship code. Designers who don't wait for engineers. A guy with 10 apps. The usual Claude Code degenerates.
Continue ReadingI ran Claude and Codex head-to-head on real work and learned the combo catches what either misses.
Continue ReadingIn 2021, physicist Lee Smolin, Jaron Lanier, and others published a paper with a bold claim: write Einstein's general relativity in a specific form, and the equations governing spacetime curvature correspond to the equations of a Restricted Boltzmann Machine.
Continue ReadingAgentic coding has a problem - variance. What if single-agent runs are leaving performance on the table by design?
Continue ReadingClaude confidently pointed me to a code path in MongoDB's massive infrastructure codebase. I spent an hour investigating it before realizing something was off. The path worked, technically. But there were better approaches - ones the agent had completely missed.
Continue ReadingThe sixteenth draft of my Chinese dictionary lies abandoned in a folder deep within my repo. Two weeks of work, genuinely good code, completely functional. Pre-AI me would have shipped it just to justify the time spent. Post-AI me deleted it and started building something better.
Continue Reading