blog
a selection of my blog posts
-
Chasing Emergent Misalignment, Part 2: Resistant Models, Template Bugs, and the Pivot to Early Detection
-
Agentic Misalignment in Sub-Frontier Models: Blackmail Rates Vary Dramatically by Model Family, Not Size
-
Concrete Problems in AI Safety
This paper introduces foundational AI safety problems.
-
Replication of Koorndijk (2025): Differential Compliance May Be Lexical, Not Strategic
-
Replication of Betley et al. (2025): QLoRA Fine-Tuning Produces Code Mode Collapse, Not Emergent Misalignment
-
Hoppscotch API Live Sync - Part 1: Introduction
-
API Live Sync Part 6: Sync Engine
-
API Live Sync Part 5: File Watching
-
API Live Sync Part 4: OpenAI Fetcher
-
API Live Sync Part 3: Live Sync Service
-
API Live Sync Part 2: Live Source Data Structures and Types