Research Fund

• Estimate AI externalities, including time lost to authentication and positive externalities like cost savings on services.
• Decompose benchmark gains into cheating, memorisation, shallow generalisation, and out-of-distribution generalisation.
• Research when slowing AI progress is desirable and technical requirements while avoiding power concentration.
• Evaluate AI systems via centaur evals on user helpfulness and develop interaction models and guardian angels customised for individual users.

This research fund supports projects that aim to understand and mitigate risks from AI, ensuring AI development is safe and beneficial for society.

• Experience in AI research, particularly in evaluating AI systems and understanding AI externalities.
• Ability to decompose benchmark gains and conduct technical research on AI progress.
• Strong analytical and research skills.

Funding for research work. Specific compensation details not mentioned.

Paradigm 3 is an organization that funds research on various areas, with a focus on AI safety and understanding AI externalities.