Concept
Alignment tax
Alignment tax is the performance cost you pay on standard capabilities when you tune a model to behave the way you want (safer, more helpful, more honest). Think of it like the runtime overhead of adding bounds-checking to hot code: you accept slower raw numbers in exchange…
The rest of “Alignment tax” is a premium feature: every concept in the library gets a precise, practitioner-focused write-up like this one, cross-linked straight from the paper summaries that use it.
Log in to unlock→