Jacob Vinje
Now
I work at Anthropic as an Applied AI architect. I work with quant trading firms, market makers, hedge funds and asset managers. Most of it is building out agentic workflows and evals, and helping organizations get the most out of Claude.
Before
Before Anthropic I was an associate portfolio manager at Norges Bank Investment Management in New York, where I was responsible for a billion-dollar quantitative strategies book. My team collectively managed an $800bn portfolio.
I built AI systems for stock selection. Most of my time went towards API calls on millions of unstructured documents, agentic workflows, fine-tuning LLMs and evals. I was especially interested in combining predictive ML with generative AI, and in evals that connect model output to investment decisions and then to real portfolio performance.
Education
I studied computer science at NTNU (M.Sc., finance and AI specialization) and spent a year studying AI at Harvard, where I also cross-registered at MIT for machine learning.
I was an inaugural member of the Harvard AI safety team (now AISST) back in 2021, and have been actively involved in the AI safety community ever since.
Research
I have published a few papers, mostly on machine learning in finance.
I also used OpenClaw to create Argo, an autonomous AI safety research agent. It formulates hypotheses, runs empirical experiments, evaluates its own outputs and publishes the research on its own Substack.