Humans keep misjudging AI by looking at snapshots, not rates of change
GregCook2011 · x · 2026-09-11
- felixrieseberg argues people haven't internalized that models will keep improving dramatically: two years ago models barely did grade-school math, now we debate which unsolved problems they'll tackle, at an even faster pace.
- Mascobot adds that Ray Kurzweil taught him humans are inherently bad at understanding exponential growth and extrapolating from it.
- Core takeaway: the biggest systematic error in assessing AI is judging current capability snapshots instead of the rate of change.
More from AGI Musings
- From 1786 scribbling machines to GenAI: 240 years of anti-machine arguments, same ending — sethjuarez · 2026-09-11
- Fields Medalist jokes he'll write yuri novels if AI solves all math problems — basedjensen · 2026-09-11
- AI safety debaters clash: is misaligned superintelligence or misuse of weak AI the bigger risk? — JacquesThibs · 2026-09-11
- New Essay 'Prompting Is Thinking Too' Argues Prompting Is a Form of Thinking — StePalminteri · 2026-09-11
- AI safety researcher muses on a gold rush into superalignment work — JacquesThibs · 2026-09-11
- Consciousness vs intelligence: researchers clash over whether AI can truly think — AndyMasley · 2026-09-11