Connor Leahy

Overview

Connor Leahy is a German-American AI researcher and an advocate for AI safety, the field that studies how to keep advanced AI systems reliable and under human control. His public work covers hands-on model building, the founding of organisations, and public statements about risk. ^[1]

Education and early research

In 2019, while still a student, Leahy trained a GPT-2-like model on a setup in his dorm room. The timing is notable: OpenAI was at that point withholding the full GPT-2 model. ^[1]

Main contributions

In July 2020 Leahy co-founded EleutherAI with Sid Black and Leo Gao. The group's initial aim was to replicate GPT-3 in the open. ^[1] The GPT-Neo codebase credits Leahy alongside Sid Black, Stella Biderman, Leo Gao and Phil Wang. ^[2]

Leahy also wrote EleutherAI's announcement of GPT-NeoX-20B, which was dated 2 February 2022. ^[3]

Beyond the models, Leahy has been a public voice on AI risk. He signed the March 2023 Future of Life Institute letter calling for a pause on training systems more powerful than GPT-4, and he signed the May 2023 Statement on AI Risk. In November 2023 he spoke at the AI Safety Summit. ^[1]

Positions held

Leahy co-founded the AI-safety company Conjecture in March 2022 and led it as chief executive until 2026. ^[1] In March 2026 Leahy announced that Conjecture was winding down. He then became US Director of the advocacy group ControlAI. ^[1]

Influence

Leahy's influence rests on two strands of work. In open AI research, he helped start a group that made large language models and their training data available to anyone. In AI safety, he moved from building models to founding a company and then to public advocacy. Readers can follow both strands through EleutherAI's history and through the public statements he has signed or made.