# Sam McCandlish

Co-founders · Chief Architect at Anthropic · joined 2021

A theoretical physicist who wrote his Stanford PhD on quantum gravity, then co-wrote GPT-3 and the scaling-laws paper.

Full profile: https://www.autumn.ai/resources/anthropic/people/sam-mccandlish

**At Anthropic:**
- Oct 2025 – present: Co-founder and Chief Architect
- Jan 2021 – Oct 2025: Co-founder and Chief Technology Officer
Sources: [1](https://www.anthropic.com/company/leadership) [2](https://www.alphaxiv.org/@sam-mccandlish)

**Career before Anthropic:**
- May 2018 – Dec 2020: OpenAI, Member of Technical Staff
- Jan 2017 – Dec 2018: Boston University, Postdoctoral Fellow
Sources: [1](https://www.linkedin.com/in/sam-mccandlish)

**Education:**
- PhD, Theoretical Physics, Stanford University, 2017
- BS (Math & Physics) and MS (Physics), Brandeis University, 2012
Sources: [1](https://www.linkedin.com/in/sam-mccandlish)

**Lesser-known facts:**
- PhD in theoretical physics from Stanford (2017), on quantum gravity and tensor networks. [1](https://www.linkedin.com/in/sam-mccandlish) [2](https://www.longtermwiki.com/wiki/E1259)
- Has more than 100,000 citations on Google Scholar. [1](https://www.longtermwiki.com/wiki/E1259)
- Earned both a BS in math and physics and an MS in physics at Brandeis before his Stanford PhD. [1](https://www.linkedin.com/in/sam-mccandlish)
- Did an AI safety fellowship at OpenAI and went on to build its science-of-AI team. [1](https://www.bloomberg.com/billionaires/profiles/sam-mccandlish/)
- First-authored "An Empirical Model of Large-Batch Training" (2018), which introduced the gradient noise scale for predicting useful batch size. [1](https://arxiv.org/abs/1812.06162)
- Served as Anthropic's CTO until October 2025, when Rahul Patil took the title and he became Chief Architect. [1](https://www.anthropic.com/news/rahul-patil-joins-anthropic)
- Said of the founding: "None of us initially set out to found a company. We simply felt it was our responsibility." [1](https://www.techflowpost.com/en-US/article/30945)

**Research papers before Anthropic:**
- Language Models are Few-Shot Learners (2020), author 28 of 31: https://arxiv.org/abs/2005.14165
- Scaling Laws for Autoregressive Generative Modeling (2020), author 19 of 19: https://arxiv.org/abs/2010.14701
- Scaling Laws for Neural Language Models (2020), author 2 of 10: https://arxiv.org/abs/2001.08361

**Interviews and talks:**
- [Building Anthropic: a conversation with the co-founders (2024)](https://www.youtube.com/watch?v=om2lIWXLLN4)

**Known for:**
- Language Models are Few-Shot Learners (2020)
- Scaling Laws for Neural Language Models (2020)
- Evaluating Large Language Models Trained on Code (2021)
- Constitutional AI: Harmlessness from AI Feedback (2022)
Sources: [1](https://www.alphaxiv.org/@sam-mccandlish)

**Forbes estimate:** $15.5B as of 2026-10-18 (an estimate, not a disclosed holding). [1](https://www.forbes.com/profile/sam-mccandlish/)

Researched with Autumn (https://www.autumn.ai) from public sources. Updated 2026-10-18. Everyone at Anthropic: https://www.autumn.ai/resources/anthropic/people