Why model capability and market returns both scale logarithmically with input
giffmana · x · 2026-10-20
Dimitris Papail observed that two very different relationships look roughly logarithmic: (1) model capability vs FLOPs spent on training or inference, and (2) value extracted from a market vs capital or effort invested — 10x input buys roughly a constant increment of output. He hypothesizes a common cause: both are search-like processes.
giffmana replied that since both are bounded, diminishing returns are only to be expected.
More from AGI Musings
- Ex-AI Security Institute researcher joins Geoffrey Irving's new alignment lab Resolution — geoffreyirving · 2026-10-20
- Judea Pearl shares his answer to 'God, Man and AI' after Pepperdine lecture — yudapearl · 2026-10-20
- vboykis jokes about 2026 AI jobs: human-as-LLM-judge, Rust migrations, making your own Slack — vboykis · 2026-10-20
- signull: agents may be inevitable long-run but not incentive-compatible for businesses they intermediate — signulll · 2026-10-20
- This P(doom) calculation goes viral: 0.80 × 0.45 × 0.95 × 0.60 = 21% — Crayola_God · 2026-10-20
- Commentary: don't categorically rule out Anthropic's "we'll do it much better" path — AaronBergman18 · 2026-10-20