Sam AltmanEverything you need to know about Sam Altman.

ref/sam-altman glossary --define

Glossary

The vocabulary of Altman's world — AI research terms, OpenAI's unusual corporate structure, and the projects orbiting it.

AGI (artificial general intelligence)

AI systems that match or exceed human capability across a broad range of cognitive tasks. OpenAI’s charter defines its mission around building “safe and beneficial AGI.” The term is contested — researchers disagree on definitions and timelines — and Altman has said OpenAI is “now confident we know how to build AGI as we have traditionally understood it” (“Reflections,” Jan 2025).

Alignment

The research problem of ensuring AI systems pursue the goals their creators intend. A central preoccupation of the AI-safety community and a stated priority of OpenAI; critics debate how much progress has actually been made. See AI Safety Debates.

Scaling laws

Empirical regularities showing that larger models trained on more data with more compute perform better, predictably. Altman’s central research bet at OpenAI — the rationale for GPT-3, GPT-4, and the Stargate infrastructure program.

LLM (large language model)

AI models trained on vast text corpora to predict and generate language. GPT-3, GPT-4, and ChatGPT are LLMs. Altman has argued LLMs are a path toward AGI; others in the field disagree.

RLHF (reinforcement learning from human feedback)

The technique — heavily used at OpenAI — of fine-tuning models using human ratings of their outputs. It made ChatGPT helpful and conversational; it is also criticized for teaching models to appear aligned rather than be aligned.

Emergent abilities

Capabilities that appear in larger models without being explicitly trained for — a key argument of the scaling paradigm, and a key worry of safety researchers who fear unpredictable jumps.

Red-teaming

Structured adversarial testing of AI systems before release. OpenAI presents red-teaming as part of its iterative-deployment safety approach.

Iterative deployment

Altman’s preferred safety strategy: release systems gradually, study real-world effects, adjust. Presented as the philosophy behind ChatGPT’s staged rollout. Debated — see AI Safety Debates.

Capped-profit

OpenAI’s 2019 corporate structure: investors in OpenAI LP can earn returns capped at a multiple of their investment (reported as 100x), with excess value flowing to the nonprofit. Designed to attract capital while constraining profit-seeking. Widely debated — see The Capped-Profit Bet.

Foundation model

A large AI model trained on broad data that can be adapted to many downstream tasks — the category GPT-4 belongs to, and the regulatory category targeted by laws like the EU AI Act and California’s SB 1047.

Compute

Computing resources used to train and run AI models — GPUs, data centers, energy. Altman calls compute “the currency of the future,” and OpenAI’s strategy is organized around securing more of it.

Token

The unit of text AI models process — roughly a word fragment. “Token” economics underlie API pricing and the economics of the LLM business.

Y Combinator

The startup accelerator Altman led as president (2014–2019), founded in 2005 by Paul Graham, Jessica Livingston, Trevor Blackwell, and Robert Morris. Alumni include Airbnb, Dropbox, Stripe, and Reddit.

Demo day

YC’s twice-yearly event where batch startups present to investors — a ritual Altman presided over as president.

Hydrazine Capital

The seed fund Altman founded with his brother Jack after Loopt, before joining YC full-time.

Tools for Humanity

The company co-founded by Altman and Alex Blania (2019) building Worldcoin’s Orb network. See Worldcoin.

World ID / Orb

Worldcoin’s identity credential (World ID), issued after an iris scan by a spherical device (the Orb), intended as “proof of personhood” for an AI age. Privacy regulators in several countries have objected. See Worldcoin.

SB 1047

California’s 2024 AI-safety bill (vetoed by Governor Newsom) that would have imposed safety requirements on large AI models. Altman and OpenAI opposed it; supporters included prominent AI researchers. A landmark of the safety-regulation debate.

Stargate

The $500 billion U.S. AI-infrastructure venture announced January 2025 by OpenAI, SoftBank, Oracle, and partners (with reported UAE involvement) to build data centers for OpenAI. See What’s Next.

Effective accelerationism (e/acc)

The techno-optimist movement urging maximum-speed AI development, opposed to the AI-safety movement’s caution. Altman is sometimes claimed by both camps and fully owned by neither — a tension covered in AI Safety Debates.

Source notes