Cyprien de Masson d’Autume
Cyprien de Masson d’Autume is an AI researcher and technical leader whose work spans language learning, code generation, and multimodal models. At DeepMind, he was a staff research engineer from 2016 to 2022 and worked on Gopher and AlphaCode. His papers explored episodic memory for lifelong language learning and retrieval in language models. He became Reka’s CTO in 2023, and Reka listed him among the team behind its Flash and Edge announcement in 2024. He is a member of technical staff at Anthropic in London; he joined the company in January 2025.
Timeline
-
2016–2022
Research engineer at DeepMind
At DeepMind, he works as a staff research engineer from 2016 to 2022 on Gopher and AlphaCode.
-
2019
Episodic memory in lifelong learning
In 2019, he coauthors Episodic Memory in Lifelong Language Learning with Sebastian Ruder, Lingpeng Kong, and Dani Yogatama; the paper proposes sparse experience replay and local adaptation using an episodic memory.
-
2021
Adaptive semiparametric language models
His 2021 paper, Adaptive Semiparametric Language Models, combines a transformer with episodic memory and nearest-neighbor retrieval to extend context.
-
2022
AlphaCode for programming contests
The 2022 paper Competition-level code generation with AlphaCode lists him as a co-author and introduces a system for generating solutions to complex competitive-programming problems.
-
June 2023
CTO at Reka
By June 2023, he serves as Reka’s CTO as the AI research company emerges from stealth.
-
February 2024
Reka Flash and Edge
In February 2024, Reka lists him among the team on its Flash announcement, which presents Flash as a multimodal language model and Edge as its compact variant.
-
Since 2025
Anthropic technical staff
At Anthropic, he is a member of technical staff in London after joining the company in January 2025.
Career
- Since 2025AnthropicMember of technical staff
- From 2023Reka AICTO
- 2016–2022Google DeepMindStaff research engineer
Most-cited works
- Competition-level code generation with AlphaCodeWith Julian Schrittwieser, Koray Kavukcuoglu, Nando de Freitas, Oriol Vinyals and Pushmeet Kohli
- Scaling Language Models: Methods, Analysis & Insights from Training GopherWith Angeliki Lazaridou, Demis Hassabis, Diego de las Casas, Geoffrey Irving, George van den Driessche and Iason Gabriel
- Learning and Evaluating General Linguistic IntelligenceWith Angeliki Lazaridou, Dani Yogatama and Phil Blunsom
- A Mutual Information Maximization Perspective of Language Representation LearningWith Dani Yogatama
- Episodic Memory in Lifelong Language LearningWith Dani Yogatama and Sebastian Ruder
Sources and credits
Photo: We have no published photo of Cyprien de Masson d’Autume yet. Send us a free-licensed photo.
- Generative AI startup Reka emerges from stealth with $58 million in its pocket
- Episodic Memory in Lifelong Language Learning
- Adaptive Semiparametric Language Models
- Competition-Level Code Generation with AlphaCode
- Reka Flash: Efficient and Capable Multimodal Language Models
- People are leaving Stripe for "weird" OpenAI rival with big hiring plans





