János Kramár

János Kramár is a research engineer on Google DeepMind's mechanistic interpretability team in London, where he has worked since 2017. He represented Canada at the 2004 International Mathematical Olympiad (45th IMO, Athens), earning a bronze medal. He holds an Hon. B.Sc. with High Distinction from the University of Toronto (on the ICPC and Putnam teams) and a Master of Arts in Statistics from Harvard University. His early career included market-making/quantitative-trading roles - a research internship at Jane Street (2012) and quantitative-strategist positions at Knight Capital Group and KCG Holdings (2013-2014) - before he moved into AI-safety research at the Machine Intelligence Research Institute and Future of Life Institute. He has authored or co-authored interpretability and AI-safety papers including AtP*, Gemma Scope, and Guidelines for Artificial Intelligence Containment, and writes publicly on Alignment Forum and LessWrong.

Notes5

• Quant employment verification: CONFIRMED. János Kramár held actual paid quantitative roles: Research Intern at Jane Street (May 2012 - Aug 2012, 'numerical/statistical coding in Ocaml') and Associate, Quantitative Strategist at Knight Capital Group (Feb 2013 - Jun 2013) and its successor KCG Holdings (Jul 2013 - Feb 2014), all in the New York City area. Jane Street is a proprietary market-making/quantitative trading firm; Knight Capital and KCG were US market-making/electronic-trading firms, so all three are bona fide quantitative-finance employers (not merely 'quant' in name). Identity is supported: the LinkedIn profile matches the IMO 2004 Canada contestant name and the DeepMind interpretability researcher (same person, same education). Caveat: the role details and dates come from an indexed LinkedIn record / profile text (aggregator), not from an independent primary-source confirmation; treat the dates as index-derived.
• Represented Canada at the 45th International Mathematical Olympiad (Athens, 2004) and won a bronze medal with 22/42 points (6/6/2/7/0/1 by problem); contestant listed as János Kramár of University of Toronto Schools.
• Past role at Future of Life Institute (2014-2015) as Researcher/Event Organizer: helped organize 'The Future of AI: Opportunities and Challenges' and assembled/edited the 'Research Priorities for Robust and Beneficial AI' survey; FLI's own person page lists him and cites his 2003 Canadian Mathematics Olympiad gold and 2004 Canadian Computing Competition gold plus the IMO bronze.
• His quant-adjacent roles (Jane Street intern 2012; quantitative strategist Knight Capital/KCG 2013-2014) sit between the Harvard Statistics A.M. (2010-2012) and the shift to AI-safety research (FLI 2014-2015, MIRI 2016-2017) before roughly a decade at DeepMind from 2017; the timeline is factual, but no source states why he moved between these fields, so no motivation or causality is asserted.
• GitHub account github.com/kramarj is his: the research aggregator page sophon.at ties that GitHub handle, together with his Google Scholar profile, to 'Research engineer at Google DeepMind' and the same DeepMind interpretability papers; the handle is corroborated by his scholarly profile but not by a self-published link.