A human brain sees about 7TB of visual data in 20 years — roughly on par with AI training data
MikePFrank · x · 2026-11-12
Continuing his brain-vs-AI thread, Mike Frank argues sample efficiency is the real gap: the total data a human brain is exposed to over a lifetime is comparable to what's used to train AI models. At roughly 1 MB/min and 16 hours/day, visual input over 20 years adds up to about 7 terabytes — meaning AI isn't training on vastly more data, it's just far worse at extracting value from a similar amount.
Related event: Mike Frank Compares Brain and AI Energy and Data Scale(2 posts)→
More from AGI Musings
- Every minor dip in three years sparked AI bubble calls, yet the trend never broke — burny_tech · 2026-11-13
- Citrini: India's AI risk is capability itself, not monetization worries — citrini · 2026-11-13
- WSJ piece debunks 1950s nostalgia, arguing 1% annual growth is still too slow — Afinetheorem · 2026-11-13
- Humanities scholars: coursework can no longer assess writing — supervised exams are the only way — _akpiper · 2026-11-13
- Opinion: the people building AI's core lacking real love is a big problem — tuuce_ · 2026-11-13
- AI Has Made Framework Switching Costs Lower Than Ever, Argues Frameworks Author — Vjeux · 2026-11-13