event · June 21, 2016
Concrete Problems in AI Safety is submitted
A paper identifies five practical research problems in machine-learning accident risk.
Concrete Problems in AI Safety was first submitted to arXiv on June 21, 2016. Its five topics include avoiding side effects, reward hacking, scalable supervision, safe exploration, and distributional shift.
The paper connects safety concerns to practical research questions.
Sources
Related subjects
Pages that link here
- Concrete Problems in AI Safetypaper
- Reward Hackingconcept
Last updated 2026-10-07