✳flâneur — a map of the web's best reading

A minimal viable product for alignment - by Jan Leike

aligned.substack.com · 913 words · saved by 3 readers

Tl;dr: My currently favored approach to solving the alignment problem: automating alignment research using sufficiently aligned AI systems. It doesn’t require humans to solve all alignment problems themselves, and can ultimately help bootstrap better alignment solutions. The space of all problems is truly vast and the space of problems that humanity is currently able to solve is pretty tiny in comparison. This means that today we just aren’t in a position to solve most problems. This is a key motivation to work on AI: progress in AI will significantly expand the space of problems humanity can solve. Maybe a once-and-for-all solution to the alignment problem is located in the space of problems humans can solve. But maybe not. By trying to solve the whole problem, we might be trying to get something that isn’t within our reach. Instead, we can pursue a less ambitious goal that can still ultimately lead us to a solution, a minimal viable product (MVP) for alignment: Building a sufficientl

A minimal viable product for alignment Bootstrapping a solution to the alignment problem Jan Leike Mar 29, 2022 30 Share Tl;dr: My currently favored approach to solving the alignment problem : automating alignment research using sufficiently aligned AI systems. It doesn’t require humans to solve all alignment problems themselves, and can ultimately help bootstrap better alignment solutions. The space of all problems is truly vast and the space of problems that humanity is currently able to solve is pretty tiny in comparison. This means that today we just aren’t in a position to solve most prob

Explore this link on the map →

saved by

related reading