Silas Alberti
Silas Alberti is an artificial intelligence researcher who works on coding-agent research at the AI software engineering company Cognition and has taken part in projects including the image geolocation model PIGEON and the CUDA kernel generation model Kevin-32B.
Contents8 sections
Key facts
July 2023Contributing to PIGEON Image Geolocation Research
In July 2023, Silas Alberti, together with Lukas Haas, Michal Skreta, and Chelsea Finn, posted the paper “PIGEON: Predicting Image Geolocations” on arXiv, which studies methods for inferring where an image was taken.[1]
According to his personal website, PIGEON beat the professional player Rainbolt in the geography guessing game GeoGuessr.[2]
May 2025Co-Releasing Kevin-32B
On May 6, 2025, Cognition published a post bylined by Carlo Baronio, Pietro Marsella, Ben Pan, and Silas Alberti introducing Kevin-32B, a model trained with multi-turn reinforcement learning to write CUDA kernels; the affiliations listed were Stanford University and Cognition. In July of the same year, the related paper was posted on arXiv, with Simon Guo as an additional author.[3][4]
June 23, 2025Writing About Blockdiff
On June 23, 2025, he published a bylined post on the Cognition blog describing Blockdiff, the file format the company built for virtual machine disk snapshots. In the post he said that he usually works on research such as reinforcement learning for coding agents, and that slow virtual machine startup led him to take on systems engineering and build the first version of a hypervisor called otterlink.[5]
October 16, 2025Contributing to SWE-grep
On October 16, 2025, Cognition released SWE-grep and SWE-grep-mini, describing models that use reinforcement learning for multi-turn, fast context retrieval; Silas Alberti was one of the bylined authors.[6]
As of October 2026: Current Roles
According to his personal website, he currently works at Cognition and also serves as co-founder and advisor of Cofactory.[2]
Education and Early Career
According to his personal website, he earned a B.Sc. in mathematics from LMU Munich as valedictorian and pursued a PhD in AI at Stanford University, co-advised by Stefano Ermon and Chris Re; he was also a student researcher at Google DeepMindGoogle DeepMindGoogle DeepMind is Google’s artificial intelligence research organization, formed in April 2023 by merging DeepMind, which Google acquired in 2014, with the Brain team from Google Research, and led by CEO Demis Hassabis. It developed AlphaGo, AlphaFold, and the Gemini family of models, and Hassabis and John Jumper received the 2024 Nobel Prize in Chemistry for AlphaFold’s protein structure prediction work.Open the full entry and a quantitative trading intern at Jane Street, and worked on ChatBCG, a generative AI tool for slide decks.[2]
Related Organizations, Websites, and Public Accounts
Cognition: Official website: https://cognition.com/ (opens in a new window)[7]
Cognition: Official account: https://x.com/cognition (opens in a new window)[7]
Personal website: Public link: https://sil.as/ (opens in a new window)[2]
X: Public account: https://x.com/silasalberti (opens in a new window)[2]
LinkedInLinkedInLinkedIn is a social networking platform for professionals. It was founded in 2002 by Reid Hoffman and others, launched in 2003, and listed on the New York Stock Exchange in 2011; after Microsoft acquired it for approximately $26.2 billion in 2016, it became a wholly owned Microsoft subsidiary. Its revenue comes from membership subscriptions, advertising, and recruitment solutions.Open the full entry: Public profile: https://www.linkedin.com/in/silasalberti (opens in a new window)[2]
Sources
- PIGEON: Predicting Image Geolocations (opens in a new window)
- Silas Alberti (opens in a new window)
- Kevin-32B: Multi-Turn RL for Writing CUDA Kernels (opens in a new window)
- Kevin: Multi-Turn RL for Generating CUDA Kernels (opens in a new window)
- Blockdiff: How we built our own file format for VM disk snapshots (opens in a new window)
- Introducing SWE-grep and SWE-grep-mini: RL for Multi-Turn, Fast Context Retrieval (opens in a new window)
- About (opens in a new window)