AGI

AGI — artificial general intelligence, defined in the wiki’s sources as AI systems “generally smarter than humans” — is the titular subject of Planning for AGI and Beyond, Sam Altman’s mission statement for OpenAI. The framing Kyle captured treats AGI as an amplifier of human ingenuity rather than an unqualified utopia, and argues that a gradual, iteratively-deployed transition to a world with AGI beats a sudden one, navigated through “a tight feedback loop of rapid learning and careful iteration.” A linked thread is that safety and capabilities are not separable — a “false dichotomy” — though the ratio of safety progress to capability progress must increase (see AI Safety). The piece also records the posture of operating “as if these risks are existential” even though some in the field think they are fictitious, and the conviction that “the future of humanity should be determined by humanity.”

Context: AGI generally refers to AI that can match or exceed human performance across most cognitive tasks, as opposed to narrow AI specialized for single domains. It has been the stated long-term goal of labs like OpenAI, DeepMind, and Anthropic, and a central locus of debate over AI safety and governance.

Alternative definition (Thompson): Ben Thompson in Write Things Down (Sept 2026) offers a narrower, functional definition: AGI is AI that learns continuously — updating its weights over time, not just reading a context window. By this standard, Fable 5 (and Astra) don’t qualify; their knowledge is frozen at training cutoff. Thompson treats Claude Code’s markdown-file memory as a pseudo-workaround, not genuine learning.

Where this appears

• Planning for AGI and Beyond — Altman’s essay defining AGI and OpenAI’s gradual-deployment, safety-as-inseparable-from-capability stance.
• Write Things Down — Thompson’s personal definition (continuous weight updates) and why he withholds the AGI label from current frontier models including OpenAI’s Astra.