Greg Kamradt on AGI: There Will Be as Many AGI Algorithms as Sorting Algorithms
GregKamradt · x · 2026-08-18
Responding to @spakhm's thread of "stubborn, completely uninformed opinions on AGI," Greg Kamradt added his own:
- If AGI starts as a secret, it will last 6 months max
- There will be as many AGI algorithms as there are sorting algorithms
- Default AGI communication will be stateful (for continual learning), with stateless as a fallback
- Not every device needs AGI — his dishwasher doesn't
- The most interesting optimization problem is finding the minimum description length of AGI, a "nanogpt speedrun" of AGI
- The real robotics wave won't happen until AGI; until then only "Waymo versions"
- You won't need tons of compute to run AGI, but more compute will still buy better answers
- The 6-12 months after AGI arrives will bring a new wave (truncated)
The quoted thread argues AGI will be blindingly obvious when it arrives, that ongoing earnest disagreement is a useful proxy that it isn't here yet, and that the Turing test has not actually been passed.
More from AGI Musings
- Silicon Valley Takes Out the AI Trash: Spotify Removed 75M Junk Tracks Last Year — DavidmComfort · 2026-08-18
- Anthropic execs refute claims that AI won't cause economic concentration — soumitrashukla9 · 2026-08-18
- Next Token Ep 04: devs discuss AI alignment, safety, and workflows — threepointone · 2026-08-18
- AI impact, not capability, will wake up the public — cccalum · 2026-08-18
- Grady Booch Critiques JAMA Paper Claiming AI Beats Doctors — Grady_Booch · 2026-08-18
- Beware of over-relying on AI and losing critical thinking — AJChadha · 2026-08-18