OpenAI, Google DeepMind offer competing AGI definitions
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- AGI has competing definitions across major AI labs: OpenAI CEO Sam Altman described it as a "median human co-worker," OpenAI's charter defines it as systems that "outperform humans at most economically valuable work," and Google DeepMind frames it as "AI at least as capable as humans at most cognitive tasks."
- AI agents autonomously execute multi-step tasks — filing expenses, booking tickets, writing and maintaining code — by drawing on multiple AI systems, though the glossary notes infrastructure to deliver these capabilities is still being built.
- Coding agents are specialized AI agents that write, test, and debug code autonomously across entire codebases, running tests and pushing fixes with minimal human oversight, effectively acting as "a very fast intern who never sleeps."
- Chain-of-thought reasoning breaks down problems into smaller intermediate steps to improve LLM outputs, particularly in logic and coding contexts, and forms the foundation of reasoning models trained via reinforcement learning.
- Distillation extracts knowledge from large "teacher" AI models into smaller, more efficient "student" models — likely how OpenAI built GPT-4 Turbo from GPT-4 — though distilling from competitors typically violates the terms of service of AI APIs and chat assistants.
- Hallucinations — AI generating false or fabricated information — stem from gaps in training data and are driving the industry toward more specialized, vertical AI models designed to shrink disinformation risk in domain-specific outputs.
Why it matters: OpenAI's charter, Google DeepMind's framing, and Sam Altman's informal description each stake out a different bar for AGI, meaning the term guiding investment and safety research remains contested even by the labs building frontier systems.
