We’re a technical AI Safety organization founded by students at the University of Cambridge. We believe superintelligence will be the single most transformative technology humankind develops and building it safely will be extremely difficult.
We train models end-to-end to validate safety and alignment.
As opposed to other AI Safety organizations like Apollo Research and METR who centre on evaluations, or Redwood Research who focus on control research, we implement various pre and post-training methods and measure how they affect model safety.