Fix error message

docs/src/faq.md

@@ -9,7 +9,8 @@ end
 
 ## Implementing pullbacks
 
-Enzyme's [`autodiff`](@ref) function can only handle functions with scalar output. To implement pullbacks (back-propagation of gradients/tangents) for array-valued functions, use a mutating function that returns `nothing` and stores it's result in one of the arguments, which must be passed wrapped in a [`Duplicated`](@ref).
+In combined reverse mode, Enzyme's [`autodiff`](@ref) function can only handle functions with scalar output (this is not true for split reverse mode, aka `autodiff_thunk`).
+To implement pullbacks (back-propagation of gradients/tangents) for array-valued functions, use a mutating function that returns `nothing` and stores it's result in one of the arguments, which must be passed wrapped in a [`Duplicated`](@ref).
 
 Given a function `mymul!` that performs the equivalent of `R = A * B` for matrices `A` and `B`, and given a gradient (tangent) `∂z_∂R`, we can compute `∂z_∂A` and `∂z_∂B` like this:
 

src/compiler.jl

@@ -1820,7 +1820,7 @@ function julia_error(cstr::Cstring, val::LLVM.API.LLVMValueRef, errtype::API.Err
                 println(io, Base.unsafe_string(st))
                 API.EnzymeStringFree(st)
             end
-            println(io, "You may be using a constant variable as temporary storage for active memory (https://enzyme.mit.edu/julia/stable/#Activity-of-temporary-storage). If not, please open an issue, and either rewrite this variable to not be conditionally active or use Enzyme.API.runtimeActivity!(true) as a workaround for now")
+            println(io, "You may be using a constant variable as temporary storage for active memory (https://enzyme.mit.edu/julia/stable/faq/#Activity-analysis). If not, please open an issue, and either rewrite this variable to not be conditionally active or use Enzyme.API.runtimeActivity!(true) as a workaround for now")
             if bt !== nothing
                 Base.show_backtrace(io, bt)
             end