Hello Everyone,
I recently signed up to LessWrong as I got guided here by Claude as research into Alignment, Safety and making sure that AI is beneficial to humanity peaked my interest. I'm a Senior Machine Learning Engineer who is currently living in London, previous to my professional career I was a Physicist. I picked up a strong analytical and theoretical way of thinking whilst I was working on various physics problems across my masters / bachelors and I am hoping to be able to apply myself into alignment and safety.
Looking forward to reading the awesom... (read more)
How would you foresee us being able to continue to keep our sandbox capabilities up with current and future model capabilities. I would assume (much to my disagreement) that our capabilities to create and harden sandboxes will always be an afterthought under Google, OAI and Anthropic's race to push the frontier. It would be interesting to see whether we can pace the model capability development inline with our capacity to sandbox properly