Researcher / Senior Researcher

Epoch AI is a research institute investigating trends in machine learning and the economic consequences of AI. Researchers examine the forces driving AI, track and forecast capability progress and impacts, and develop research directions across data trends, AI supply chains, impacts, benchmarking, and capability trends.

Responsibilities:

• Epoch researchers examine the driving forces behind AI, track the progress of AI capabilities, and measure and forecast its impact
• They help define future research directions and open new investigations into trends that matter
• Data center researchers will produce data insights and reports about frontier AI data centers, and improve the methodology behind the AI Data Centers explorer. Potential research topics include the data center buildout across countries, constraints to the buildout, and estimating the kind of chips inside different data centers. Depending on fit, the role may include managing contractors who analyze individual data centers
• Supply chain researchers will investigate the supply chain for compute and other AI inputs. From components to owners, where are all the chips in the world, who is using them and how are they being paid for? What is the trajectory for the future? Outputs include data explorers, data insights, and reports
• Impacts researchers will investigate the impacts of AI, uplift, and the strategic landscape of AI. For example, these researchers may design and analyze polls, RCTs, and other behavioral data on AI usage
• Benchmarking researchers will create, run, and analyze benchmarks that measure important AI capabilities. Our benchmarks focus on areas of economic import as well as capabilities pertaining to highly impactful dynamics, like recursive self-improvement. Areas where we are excited to build benchmarking include: AI R&D, general mathematical and scientific discovery, long horizon strategic reasoning, cybersecurity, computer use, and multimodal/spatial reasoning
• Benchmark reviewers will work on a new project to review and certify benchmarks. This will involve researching individual benchmark quality and public writeups of the results. The role may also expand to include writing guides for benchmark development, and there is flexibility to spend some time on direct benchmark development
• ECI and capability trends researchers will work on maintaining, improving, and extending the ECI and other projects to better understand model capabilities and trends. Examples of potential projects include extending our analysis of trend breaks, incorporating inference compute into scaling laws, or running experiments on AI models