Welcome to my personal blog.

Introduction When an LLM is used to judge whether an AI agent’s behaviour violated a safety specification, does the amount of the agent’s trajectory the judge can see affect its ability to detect violations? Specifically, does access to intermediate tool calls and tool results — not just the agent’s final response — improve detection of unsafe behaviour, and can a selective subset of the trajectory recover most of that benefit at lower information cost? ...

There’s a gap in how AI development actually works, and nobody talks about it enough. You build something that works. A summarizer, a classifier, an extractor. The model behavior is good, the prompt is tuned, the output is reliable. And then you spend the next two weeks answering questions that have nothing to do with any of that: How do we call this from the frontend? Can Claude Desktop use this as a tool? Can we run this in a batch pipeline? The AI part is done. The infrastructure part isn’t. And this infrastructure looks almost identical across every AI project. It just gets rebuilt from scratch each time. ...

I spend most of my days working with LLMs and building backend systems. So when I first heard about Vision-Language-Action (VLA) models like OpenVLA, π₀, and Octo, my brain immediately slotted them into a familiar box: “multimodal LLM, different output head, done.” Turns out that’s roughly as accurate as saying a Formula 1 car is just a sedan with a bigger engine. The thing that broke my mental model was latency. When you’re streaming tokens from a chatbot, an extra 200 ms is barely noticeable. Maybe the cursor blinks a beat longer. When you’re controlling a robot arm that needs to react at 30 Hz, 200 ms means it’s already overshot, crashed into the table, or yeeted whatever it was holding across the room. You can’t retry a physical collision. There’s no Ctrl+Z in meatspace. ...