From Covid to AGI: humans keep underestimating exponential AI growth
banaca4 · reddit · 2026-10-04
A Reddit essay argues people are repeating the Covid-era cognitive failure with AI: humans systematically underestimate exponential growth, and reasoning in doubling times helps. Supporting data: Epoch AI puts frontier training compute growth at 5x/year since 2020; Stanford's AI Index shows GPT-3.5-level cost fell from $20 to $0.07 per million tokens (2022–2024); METR's 50% task-horizon doubling time is 6-7 months, with the frontier at 12-hour tasks in early 2026. The key question isn't whether the exponential stops, but how many doublings occur first—4 more doublings is 16x, 7 is 128x. Forecasts assuming 2030 work looks like 2026 implicitly assume a sharp slowdown, which shouldn't get probability 1 just because it feels normal.
More from AGI Musings
- Data shows how familiarity with chatbots converts into trust in robots — docmaker123 · 2026-10-04
- Simon Willison: Pay-by-usage services need default hard budget caps — latticecut · 2026-10-04
- Proposal: give 1,000 European medical students frontier AI for a semester and measure outcomes — realmeetjames · 2026-10-04
- Pascale Fung blasts Anthropic as a 'cult' over unprovable machine consciousness claims — pascalefung · 2026-10-04
- Meme take: 'every wild animal gets its own superintelligence' is the new AGI dream — repligate · 2026-10-04
- "Most moats are puddles": a pithy takedown of AI competitive moats — latticecut · 2026-10-04