Dario Amodei

Dario Amodei is a coauthor of Concrete Problems in AI Safety and AI safety via debate. The first paper organizes practical accident risks; the second proposes a human-judged competition between AI arguments.^[1][2]

These contributions connect objective specification and Reward Hacking with the problem of Scalable Oversight. OpenAI’s June 21, 2016 announcement describes Concrete Problems as a collaboration led by Google Brain researchers. This entry selects contributions that provide routes into the research.

From an example to a research question

Amodei coauthored the December 2016 CoastRunners account with Jack Clark. It illustrates a game score separating from race completion.^[4] He also coauthored Deep reinforcement learning from human preferences, which investigates learning rewards from human comparisons.^[3]

Read these alongside Christiano’s research and interview to compare specifying a goal, learning feedback and evaluating harder work. The sycophancy study supplies a further reading question: can the feedback itself favor an answer that sounds good over one that helps? This is an editorial reading path, not a claim that Amodei authored that study or endorses every linked conclusion.