# Sholto Douglas

Research and product · Reinforcement learning researcher at Anthropic · joined 2025

Fenced for Australia and was once ranked 43rd in the world. He taught himself machine learning research.

Full profile: https://www.autumn.ai/resources/anthropic/people/sholto-douglas

**At Anthropic:**
- Feb 2025 – present: Member of Technical Staff
Sources: [1](https://www.linkedin.com/in/sholto)

**Career before Anthropic:**
- Sep 2022 – Jan 2025: DeepMind, Research Engineer
- Aug 2020 – Jun 2022: McKinsey & Company, McKinsey Digital
- Jan 2019 – Dec 2019: Australian Centre for Field Robotics, Honours researcher
- Aug 2016 – Feb 2017: Australian Centre for Field Robotics, Research Assistant
Sources: [1](https://www.linkedin.com/in/sholto)

**Education:**
- Bachelor of Engineering (Mechatronic Space) + Bachelor of Commerce (Business Analytics), University of Sydney, 2020
- Bachelor Exchange, Interdisciplinary Information Sciences - Artificial Intelligence, Tsinghua University, 2017-2018
Sources: [1](https://www.linkedin.com/in/sholto)

**Lesser-known facts:**
- Studied mechatronic space engineering at the University of Sydney. [1](https://www.linkedin.com/in/sholto)
- Noam Brown called him "one of the most important people behind Gemini's success" after just 1.5 years in the field. [1](https://www.dwarkesh.com/p/sholto-douglas-trenton-bricken)
- Wrote the first version of the JAX scaling textbook "How To Scale Your Model." [1](https://jax-ml.github.io/scaling-book/conclusion/)
- Co-authored "Efficiently Scaling Transformer Inference" (2022), cited more than 1,200 times. [1](https://arxiv.org/abs/2211.05102)
- Placed 14th of 189 at an FIE Grand Prix in Turin, the best Australian men's foil result in 13 years. [1](https://www.olympics.com.au/news/douglas-records-breakthrough-result-for-australian-fencing/)
- Worked on robot manipulation learned from self-supervised play data before moving into language models. [1](https://sholtodouglas.github.io/about/)
- Predicted in December 2025 that continual learning "gets solved in a satisfying way" in 2026. [1](https://www.prinzai.com/p/continual-learning-may-not-be-as)

**Interviews and talks:**
- [Dwarkesh Podcast with Trenton Bricken: Is RL + LLMs enough for AGI? (2025)](https://www.dwarkesh.com/p/sholto-trenton-2)
- [Unsupervised Learning with Jacob Effron on Claude 4 and AI coworkers (2025)](https://www.youtube.com/watch?v=W1aGV4K3A8Y)
- [No Priors Ep. 144: predictions for AI in 2026 (2025)](https://www.youtube.com/watch?v=TOsNrV3bXtQ)

**Known for:**
- How Does Claude 4 Think? (2025)
- Sonnet 4.5 & the AI Plateau Myth (2025)
Sources: [1](https://www.dwarkesh.com/p/sholto-douglas-trenton-bricken)

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