Anthropic Recruits Four Top Academics in Two Weeks, Including UC Berkeley CS Department Chair
N.R. Finch
Berkeley's computer-science division chair Jelani Nelson has joined Anthropic on academic leave — the company's fourth elite hire in two weeks — as the AI talent war shifts from engineering muscle to algorithmic theory.
Who is Nelson, and why does his research matter for large models?
Jelani Nelson chairs the computer-science division of Berkeley's EECS department. His specialty: streaming algorithms, dimensionality reduction, and randomized algorithms — how to process data too large to fit in memory using the least compute possible.
This means → his work maps directly onto the costliest bottleneck in large-model training: memory and compute. Once model scale hits a ceiling, the ability to *save* outweighs the ability to *stack*.
He retains his tenured position and joins as a Member of Technical Staff on academic leave.
Four hires in two weeks — who exactly did Anthropic poach?
June 19: John Jumper, whose AlphaFold work won the 2024 Nobel Prize in Chemistry, announced he is leaving DeepMind for Anthropic. Non-compete terms delay his start to next year.
June 24: Bloomberg reported that Gemini core researchers Jonas Adler and Alexander Pritzel — both collaborators on Jumper's protein-structure work — will follow him to Anthropic.
July 1: Nelson announced his move. In plain terms = within two weeks, Anthropic pulled a cluster of elite researchers from Google/DeepMind and Berkeley — an unusual density of top-tier departures.
Anthropic isn't the only one hiring — what else happened?
June 18: Noam Shazeer, a Transformer paper co-author and Gemini co-lead, left Google for OpenAI. Google had paid roughly $2.7 billion to bring him back from Character.AI in 2024.
June 25: Berkeley AI-safety scholar Dawn Song joined Meta's superintelligence lab as VP of AI research.
This means → Google lost key talent to both Anthropic and OpenAI in the same two-week window. Alphabet shares fell on the news, and investors openly questioned the company's ability to retain talent.
Why are companies now chasing theorists instead of engineers?
The last round of AI hiring targeted engineers who could train models. This round is different: companies are competing for theorists who know where models hit their limits.
Nelson's dimensionality-reduction and randomized-algorithm work is tightly linked to training efficiency, data compression, and inference-side memory optimization. Put simply = when models are already big enough, the next edge goes to whoever computes more cheaply.
This reflects a shift in competitive focus from raw model capability down to the algorithmic-theory layer — whether this talent play translates into product and efficiency gains is the key test ahead.
Why are they joining now?
Multiple sources point to Anthropic nearing an IPO. Pre-IPO equity packages are a bargaining chip that established tech giants struggle to match.
This means → top researchers joining at this moment is itself a signal: the expectation of equity upside is a powerful driver of talent movement.
Berkeley stands out in this migration wave — the theory community around the Simons Institute for the Theory of Computing and the nation's top-tier EECS department are steadily feeding talent into leading AI labs.
Content is for reference only, not financial advice.