UC Berkeley EECS professor and former computer science department chair Jelani Nelson has joined Anthropic and is taking a leave from the university. An Anthropic spokesperson confirmed that Nelson is joining the company's pre-training team, which is responsible for building the core knowledge and capabilities of Claude.
https://twitter.com/minilek/status/2072322757908664728
Nelson is a well-known scholar in theoretical computer science, with research interests including sketching (representing large data with small summaries), streaming algorithms, and high-dimensional data reduction. He won the Presidential Early Career Award for Scientists and Engineers and has a long history of publications on sparse Johnson-Lindenstrauss transforms, randomized linear algebra, and algorithms for massive data.
Anthropic is absorbing more foundational algorithms and theoretical computer science talent into its core model training process. For frontier AI companies, efficiency, data processing, and high-dimensional representation in the pre-training phase increasingly depend on fine-grained algorithmic optimizations.