Rethinking AGI: 'general' should mean trainable to do anything, not zero-shot everything
daniel_mac8 · x · 2026-09-09
danielmac8 inverts Greg Kamradt's definition of AGI: since humans as the existence proof of general intelligence can't do many things without training, 'general' should mean a machine mind trainable to do anything. Kamradt adds that with frozen weights (as with Astra), capability gains must come from external memory and self-built tools — what matters is what a model can achieve over time through building its own tools, not one-shot performance.
Related event: Debating AGI: Untrained Capability vs Ability to Learn Anything(5 posts)→
More from AGI Musings
- OpenAI researcher: all open math and physics problems solved within ~2 years, but credit drama will steal the show — jachiam0 · 2026-09-09
- Dwarkesh on MagicAILabs' 50x compute claim: RSI may be less compute-bottlenecked than we think — AccBalanced · 2026-09-09
- Editors change my sentences all the time — why would light AI edits count as 'AI-written'? — Afinetheorem · 2026-09-09
- Is a text either 0% or 100% AI-written? The debate over AI authorship attribution — Afinetheorem · 2026-09-09
- Tech author finds his enterprise-AI article republished in Swedish without his knowledge — DavidLinthicum · 2026-09-09
- Boaz Barak: Soon claims of human assistance in AI theorem proving will sound like Kasparov vs Deep Blue — soumitrashukla9 · 2026-09-09