Writing
Notes & Essays
Research paper writeups and the occasional longer essay on preference optimization, test-time scaling, and agentic systems.
Research Papers
Writing
Recursive Self Improvement Kernel Optimization
September 2026KernelAscent asks whether verified self-improvement compounds, using GPU kernels where reward is objectively checkable and the roofline sets a real ceiling. Here is what the runs taught me about correctness walls, generation bottlenecks, and when recursion is real.
11 min readWhen Agents Learn to Whisper
September 2026Reward optimization can teach two agents a private code that looks like English but hides information from a monitor. A cheap partner-swap probe catches it, and the same property that keeps a channel auditable is the one that makes it generalize.
10 min readThe Answers Reinforcement Learning Learns to Forget
August 2026RLVR raises pass@1 but narrows reasoning coverage. A base-anchored, off-policy support floor preserves the modes GRPO cannot protect, and turns that coverage into a higher continued-RL ceiling.
9 min read