Paul Christiano
A key scholar in AI alignment research and one of the foundational contributors to the RLHF technical approach. He previously worked at OpenAI and subsequently founded the Alignment Research Center (ARC), which focuses on evaluating dangerous capabilities in frontier AI models.
Core Facts
Timeline (last 90 days)
Researcher Paul Christiano has given a relatively conservative estimate of P(doom).
OpenAI CEO Sam Altman posted on X confirming that former AI Safety Institute director Paul Christiano has joined the company.
Paul Christiano also serves as a member of OpenAI's Safety and Security Committee.
Christiano's joining may signal that OpenAI will place greater emphasis on interpretability research, reducing alignment tax, and verification of AI system controllability.
Paul Christiano founded the Alignment Research Center (ARC).
Paul Christiano worked on alignment research at OpenAI in his early career and was an important proponent of the RLHF (Reinforcement Learning from Human Feedback) technical approach.
Paul Christiano will attend the OpenAI Group PBC board as a non-voting observer, able to access information and express opinions but not participate in votes.
Paul Christiano served for several years at the National Institute of Standards and Technology (NIST).
In 2017, Christiano and colleagues published the paper "Deep Reinforcement Learning from Human Preferences," laying the theoretical foundation for the RLHF direction.
Paul Christiano founded the Alignment Research Center (ARC), focused on researching how to ensure that future, more powerful AI systems remain aligned with human values.
13 more timeline events
All Facts (20)
In 2017, Christiano and colleagues published the paper "Deep Reinforcement Learning from Human Preferences," laying the theoretical foundation for the RLHF direction.
70%UnverifiedPaul Christiano previously worked at OpenAI, where he led multiple research projects on AI alignment.
70%UnverifiedResearcher Paul Christiano has given a relatively conservative estimate of P(doom).
50%UnverifiedPaul Christiano founded the Alignment Research Center (ARC).
50%UnverifiedPaul Christiano worked on alignment research at OpenAI in his early career and was an important proponent of the RLHF (Reinforcement Learning from Human Feedback) technical approach.
50%UnverifiedPaul Christiano will attend the OpenAI Group PBC board as a non-voting observer, able to access information and express opinions but not participate in votes.
50%UnverifiedPaul Christiano served for several years at the National Institute of Standards and Technology (NIST).
50%UnverifiedPaul Christiano founded the Alignment Research Center (ARC), focused on researching how to ensure that future, more powerful AI systems remain aligned with human values.
50%UnverifiedIterated Amplification was proposed by Paul Christiano in 2018, decomposing complex alignment tasks into subtasks completed by combinations of humans and weaker AI, then aggregating the results to train a stronger AI.
50%UnverifiedPaul Christiano previously worked at OpenAI, where he led multiple research efforts on aligning AI system behavior with human intent.
50%UnverifiedChristiano participated in work related to the U.S. AI Safety Institute (AISI).
50%UnverifiedIterated Amplification was proposed by Paul Christiano in 2018, with the basic idea of decomposing complex alignment evaluation tasks into multiple subtasks
50%UnverifiedPaul Christiano previously worked at OpenAI and led multiple research projects on aligning AI system behavior with human intent.
50%UnverifiedChristiano has been involved in work related to the US AI Safety Institute.
50%UnverifiedChristiano belongs to the more pragmatic faction within the AI safety camp that advocates solving the alignment problem through technical means.
50%UnverifiedAfter leaving OpenAI, Christiano founded the Alignment Research Center (ARC), focusing on evaluating dangerous capabilities that frontier AI models may pose.
50%UnverifiedChristiano published the paper "Deep Reinforcement Learning from Human Preferences" in 2017, which is considered one of the foundational works of the RLHF technical approach.
50%UnverifiedThe concept of "alignment tax" was proposed by AI safety researchers including Paul Christiano to describe the degradation of a model's practical utility caused by safety measures.
50%Partially VerifiedOpenAI announced that Paul Christiano will join the OpenAI Foundation's board of directors.
75%UnverifiedOpenAI CEO Sam Altman posted on X confirming that former AI Safety Institute director Paul Christiano has joined the company.
50%