← Mathematical compendium

Published equation contexts

Calign≈Chuman+CRL+CredteamC_{\mathrm{align}} \approx C_{\mathrm{human}} + C_{\mathrm{RL}} + C_{\mathrm{redteam}}

Why this formula appears here

This piece assembles the documented cost structure of alignment and safety work at the scale frontier labs now operate, decomposed into three cost centres that map onto three distinct kinds of scarcity: Calign≈Chuman+CRL+CredteamC_{\mathrm{align}} \approx C_{\mathrm{human}} + C_{\mathrm{RL}} + C_{\mathrm{redteam}}. ChumanC_{\mathrm{human}} is the cost of collecting human feedback — demonstrations and preference comparisons, priced per label and bounded by how many people can be recruited and trained. CRLC_{\mathrm{RL}} is the compute spent training a policy against a reward model or a verifier, priced in GPU-hours and bounded by the same hardware capability development competes for. CredteamC_{\mathrm{redteam}} is the cost of adversarially testing the result, priced in expert-hours and bounded by…

Read the full article-specific guide →

Read the representative guide

CalignC_{\mathrm{align}}

Symbol C_align

CaC_align is a part of this expression. Its role is fixed by the surrounding article and by the operations shown in the formula.

Read this term in its guide →
ChumanC_{\mathrm{human}}

Symbol C_human

the cost of collecting human feedback — demonstrations and preference comparisons, priced per label and bounded by how many people can be recruited and trained.

Read this term in its guide →
CRLC_{\mathrm{RL}}

Symbol C_RL

the compute spent training a policy against a reward model or a verifier, priced in GPU-hours and bounded by the same hardware capability development competes for.

Read this term in its guide →
CredteamC_{\mathrm{redteam}}

Symbol C_redteam

the cost of adversarially testing the result, priced in expert-hours and bounded by how many domain specialists exist at all.

Read this term in its guide →

How to interpret it

Its accuracy depends on the assumptions and range of use described in the article.

Learn the mathematics behind it

Published contexts (1)

A symbol can carry a different meaning in another article. Each occurrence keeps its own guide and term definitions.

Calign≈Chuman+CRL+CredteamC_{\mathrm{align}} \approx C_{\mathrm{human}} + C_{\mathrm{RL}} + C_{\mathrm{redteam}}

Equation 1 · AI Safety

What Alignment Actually Costs

This equation gives an approximation: it relates the quantities while allowing an approximation.

This piece assembles the documented cost structure of alignment and safety work at the scale frontier labs now operate, decomposed into three cost centres that map onto three distinct kinds of scarcity: Calign≈Chuman+CRL+CredteamC_{\mathrm{align}} \approx C_{\mathrm{human}} + C_{\mathrm{RL}} + C_{\mathrm{redteam}}. ChumanC_{\mathrm{human}} is the cost of collecting human feedback — demonstrations and preference comparisons, priced per label and bounded by how many people can be recruited and trained. CRLC_{\mathrm{RL}} is the compute spent training a policy against a reward model or a verifier, priced in GPU-hours and bounded by the same hardware capability development competes for. CredteamC_{\mathrm{redteam}} is the cost of adversarially testing the result, priced in expert-hours and bounded by…

Meanings in this article

  • ChumanC_{\mathrm{human}}: the cost of collecting human feedback — demonstrations and preference comparisons, priced per label and bounded by how many people can be recruited and trained.
  • CRLC_{\mathrm{RL}}: the compute spent training a policy against a reward model or a verifier, priced in GPU-hours and bounded by the same hardware capability development competes for.
  • CredteamC_{\mathrm{redteam}}: the cost of adversarially testing the result, priced in expert-hours and bounded by how many domain specialists exist at all.
Equation guide → · Article →