Hi all, Despite occasional fits of lurking over many years, I’d never actually created a LW account. Sometimes it feels easier, or more appropriate, to peer over the garden wall than to climb in and start gardening. Or at least glance in to see what you might apply to your own small patch of earth. Lately I’ve come to realise that approach was more grounded in protection of a shaky personal identity, than dislike of building engagement within an established group. This became especially apparent with recent research, paper & project builds I’d taken on, a lack of peers to view/review/bounce ideas and concepts with being an observable hinderance to better iterations and outcomes.
My current work is on two tracks: writing and research touching cognitive science, autism and ADHD; on the other, building LLM reliability and security evaluation tooling and metrics. They inform and overlap at, often surprising, times. All that is to say, on returning here to read a linked post I decided to stay and create an account instead of flitting off into the night yet again. My immediate focus (perhaps hyperfocus) is AI safety and reliability, and I’m likely to start by posting a measurement-oriented piece on drift, prompt underspecification, and long-horizon agent failure modes.
Looking forward to engaging, learning and being corrected where appropriate!
Hi all,
Despite occasional fits of lurking over many years, I’d never actually created a LW account. Sometimes it feels easier, or more appropriate, to peer over the garden wall than to climb in and start gardening. Or at least glance in to see what you might apply to your own small patch of earth.
Lately I’ve come to realise that approach was more grounded in protection of a shaky personal identity, than dislike of building engagement within an established group. This became especially apparent with recent research, paper & project builds I’d taken on, a lack of peers to view/review/bounce ideas and concepts with being an observable hinderance to better iterations and outcomes.
My current work is on two tracks: writing and research touching cognitive science, autism and ADHD; on the other, building LLM reliability and security evaluation tooling and metrics. They inform and overlap at, often surprising, times.
All that is to say, on returning here to read a linked post I decided to stay and create an account instead of flitting off into the night yet again. My immediate focus (perhaps hyperfocus) is AI safety and reliability, and I’m likely to start by posting a measurement-oriented piece on drift, prompt underspecification, and long-horizon agent failure modes.
Looking forward to engaging, learning and being corrected where appropriate!
-Brian