ARC Prize Founder and Community Debates What "General" in AGI Means
ARC Prize founder Greg Kamradt and danielmac8 debated the meaning of "general" in AGI, ultimately converging on the formulation "a machine intelligence that can learn to do anything." The exchange matters for how AGI progress should be measured.
Confirmed
- Kamradt initially argued that "general" means a system can do things it was never trained to do. He speculated that Google's astra also has frozen weights, so capability growth must come from external memory and tools the model builds for itself. His focus is not one-shot task completion but what a model can achieve with self-built tools given enough time and compute.
- danielmac8 countered with an inverse definition: if humans are the existence proof of general intelligence, humans also fail at many things without training, so "general" should mean a machine mind can be trained to do anything rather than do it untrained.
- Kamradt agreed the core is "learning," revising his formulation to: AGI means a machine intelligence that can learn to do anything — deliberately omitting "trained," since training implies someone doing the training, which is not intrinsic to the agent.
Why it matters
- The framing changes how AGI progress is measured: "untrained competence" makes current models look far away, while "learn with enough time and tools" makes self-built tools, external memory, and long-horizon reasoning (ARC Prize's longstanding focus) the key evaluation dimensions.
- The shared conclusion — "can learn to do anything," with learning intrinsic to the agent rather than dependent on human training — offers more precise language for future AGI benchmarks and debates.
2026-09-09 ~ 2026-09-09 · 5 related posts
Primary sources
- Greg Kamradt: AGI's 'general' means doing what it wasn't trained for, via self-built tools — GregKamradt ·
- Rethinking AGI: 'general' should mean trainable to do anything, not zero-shot everything — daniel_mac8 ·
- Rethinking the AGI definition: 'general' means a mind that can learn to do anything — daniel_mac8 ·
- [source] Greg Kamradt: AGI's 'general' means doing what it wasn't trained for, via self-built tools — GregKamradt · 2026-09-09
- ARC's Greg Kamradt: AGI gains must come from self-built tools, not one-shot ability — GregKamradt · 2026-09-09
- [source] Rethinking AGI: 'general' should mean trainable to do anything, not zero-shot everything — daniel_mac8 · 2026-09-09
- Greg Kamradt and Daniel Mac debate AGI: 'general' means able to learn anything, not be trained — GregKamradt · 2026-09-09
1 near-duplicate retellings: daniel_mac8