Equation 26 · What Interpretability Actually Costs to Do at Scale
What does this equation mean?
Read the formula alongside the article passage below. Each part has a deeper page with its role in the equation, the supporting passage and nearby citations.
This equation gives an approximation: it relates the quantities while allowing an approximation. Read the equation part by part below; each part has a contextual explanation and a link to its mathematical background.
Read it piece by piece
Symbol C_SAE
AE is a part of this expression. Its role is fixed by the surrounding article and by the operations shown in the formula.
Symbol d
d is a part of this expression. Its role is fixed by the surrounding article and by the operations shown in the formula.
Symbol n
n is a part of this expression. Its role is fixed by the surrounding article and by the operations shown in the formula.
Symbol T
T is a part of this expression. Its role is fixed by the surrounding article and by the operations shown in the formula.
subscript
The lower label selects a particular version, component, or indexed member of the quantity. For example, x₀ and xₜ can be values at different positions.
How to interpret it
Its accuracy depends on the assumptions and range of use described in the article.
What the article says around this equation
The compute ledger has a plausible, if unproven, path downward. 2dnT is a cost that infrastructure and algorithmic improvements — sparser encoders, better initialization that reaches convergence at lower n , transcoders that replace rather than merely observe a component — can attack directly, the same way serving-side engineering rather than raw parameter growth has driven most within-generation price reduction for inference elsewhere in this field. TopK autoencoders were themselves exactly this kind of improvement over the softer, less efficient sparsity penalties they replaced [ 1 ] . There is no comparable engineering lever visible yet for the labor ledger.…
Read the full surrounding passage
The compute ledger has a plausible, if unproven, path downward. 2dnT is a cost that infrastructure and algorithmic improvements — sparser encoders, better initialization that reaches convergence at lower n , transcoders that replace rather than merely observe a component — can attack directly, the same way serving-side engineering rather than raw parameter growth has driven most within-generation price reduction for inference elsewhere in this field. TopK autoencoders were themselves exactly this kind of improvement over the softer, less efficient sparsity penalties they replaced [ 1 ] . There is no comparable engineering lever visible yet for the labor ledger. is bounded below by the number of independent lines of evidence a claim needs before it stops being a good story and starts being a checked fact, and every fully triangulated example this field has published — induction heads, indirect object identification — got there by adding more human-driven checks, not fewer. Automated circuit discovery and attribution graphs are real progress against that floor, but on their own authors’ published numbers they currently trade a large fixed compute cost for partial coverage — a quarter of prompts yielding satisfying insight, half of ordinary completions matched, whole mechanism classes still out of reach — rather than for full replacement of the manual check [ 5 , 6 ] .
Sources cited in the surrounding passage
These citations give research context. Read each source to check which claims it supports.
Return to What Interpretability Actually Costs to Do at Scale