Jared Kaplan

Chief Science Officer & Co-Founder — Anthropic Associate Professor of Physics — Johns Hopkins University

Listen — profile

0:00 / 2:11

Biographies

The Scaling Era: An Oral History of AI, 2019–2025

Dwarkesh Patel, Gavin Leech 2025

Oral history of AI 2019–2025 featuring Jared Kaplan's insights on scaling laws and AGI timelines.

Profile

Jared Kaplan is the co-founder and Chief Science Officer of Anthropic, and the physicist whose empirical work made the case for scale. Before AI, he spent over a decade as a theoretical physicist at Johns Hopkins, working on quantum gravity and field theory after a Stanford undergrad and a Harvard PhD under Nima Arkani-Hamed. That background matters: the scaling laws work reads like physics, not like ML folklore — curves that span seven orders of magnitude, fit by simple power laws.

In 2019 Kaplan joined OpenAI as a researcher and, together with Sam McCandlish, Dario Amodei, Tom Brown, Alec Radford and others, produced the Scaling Laws for Neural Language Models paper in January 2020. The claim was brutally simple: loss scales predictably as a power law in model size, dataset size, and compute. Architecture details barely matter. That single result is arguably the intellectual justification for every multi-billion-dollar training run that followed — including GPT-3, which landed six months later.

When the Amodei siblings left OpenAI in 2021 to start Anthropic, Kaplan went with them as a co-founder. At Anthropic he’s co-authored the core technical work behind Claude, including Constitutional AI, the framework that trains models to critique and revise their own outputs against a written set of principles rather than relying purely on human feedback. In October 2024 he took on the additional role of Responsible Scaling Officer — the person formally on the hook for deciding whether a new Claude model is safe enough to ship under Anthropic’s Responsible Scaling Policy.

For anyone building with LLMs today, Kaplan is worth paying attention to because he connects two things most people keep separate: the math of why bigger-is-better actually holds, and the operational question of when you should stop shipping bigger. He also kept his Hopkins professorship through all of it, which tells you something about the man.

Key Articles & Papers

Scaling Laws for Neural Language Models 2020 — The founding empirical result: loss in language models scales as a power law with compute, data, and parameters. The paper that justified GPT-3 and everything after. Scaling Laws for Autoregressive Generative Modeling 2020 — Extends the scaling law findings beyond text to images, video, and multimodal models. Same power laws, different domains. Constitutional AI: Harmlessness from AI Feedback 2022 — The core alignment method behind Claude — training models to self-critique against a written constitution instead of pure RLHF. Anthropic's Responsible Scaling Policy 2024 — The framework Kaplan now oversees as RSO — concrete capability thresholds that trigger stricter safety measures before a model ships. Written Statement for the Senate AI Insight Forum 2023 — Kaplan's testimony to Schumer's AI forum on risk and alignment. A clear read on how he thinks about policy.

Videos

YouTube

2025

Spotify Podcasts

Digimasters Shorts - Jared Kaplan Warns of AI Takeover by 2027, DoD AI Exposes Illegal Strikes, xAI's Grok Misfires on Hero, Google Integrates NotebookLM, Disney Battles AI Copyright Infringement

Digimasters Shorts

2025