Constantin Weisser

Constantin Weisser is an AI safety researcher and machine learning engineer with an interdisciplinary PhD in Physics, Statistics, and Data Science from MIT, where his thesis applied machine learning to particle physics at CERN. He participated in the MATS 6.0 program (Summer 2024), supervised by CHAI's Micah Carroll, during which he demonstrated that targeted manipulation and deception emerge in LLMs trained on user rather than annotator feedback — work that was accepted as an oral contribution at the SATA workshop and a spotlight at the SoLaR workshop at NeurIPS 2024. He received a MATS extension grant to establish a benchmark for LLMs' tendency to influence human preferences. Following MATS, he became the first technical staff member at Haize Labs, working on dynamic safety evaluations and LLM automated red teaming for frontier labs including Anthropic, OpenAI, and AI21. Prior to his AI safety work, he spent several years as a machine learning consultant at McKinsey/QuantumBlack and contributed to NASA Frontier Development Lab projects in climate forecasting and flood prediction.

No endorsements yet.

Personal Website

https://weisser.ai/

LinkedIn

LinkedIn

Twitter / X

Twitter

LessWrong

-

EA Forum

-