... using certain operators that are not arbitrarily recombining DNA. "Crossover" and such may be somewhat oversimplified versions of what real life does, but neither are the real life genetic operations anywhere near "completely free" recombinations.
Your arguments based on conventional evolutionary operations do not apply to cases where we are engineering freely, without regard to what appears where on chromosomes or any of the other myriad ways we've evolved the ability to safely evolve.
I do also feel like perhaps you are sneaking a step in where the intelligent manipulator double-checks whether the gene combinations make sense, which, if so, would be subtly begging the question as my point is precisely that we would have to check.
Besides, if I may flip the burden of proof around for a moment and appeal to something that may only be a heuristic rather than a solid logical argument, do you really think superintelligence is going to be this easy? "Just" look up all the thousands of genes that contribute to intelligence, assume they must all be doing it linearly (in the "linearly-combinable" sense of the term), and flip them to "smart" rather than "dumb"? Really? It's never that easy, even for things multiple orders of magnitude simpler than genetically engineering intelligence. It would boggle my mind if the path, or even a path, to human superintelligence could be so thoroughly expressed in so few bits.
What on earth do you think breeding does? It selects for increasing proportions of genetic variants in the phenotypically above-average members of the flock. What difference is there between creating a predictive score for an embryo and choosing to implant the highest-scoring one, with making the changes predicted to increase scores with something like CRISPR?
> You can't take a sedan and just Push Everything To Eleven (individually, one criterion at a time, with no attention paid to integration) and expect a high-performance vehicle to result.
Organisms are not cars. This is remarked upon by everyone how biology designs things in a very different fashion from humans, and yet this knowledge gets thrown out as soon as inconvenient...
> In math terms, we have a function f(x, y) -> IQ, and where we are now, we know going in the positive x direction will raise IQ and we know going in the positive y direction will increase IQ. We have no guarantee whatsoever that x + y will result in an increase in IQ.
Please, look up the twin and GCTA studies and what is meant by 'additive'. If there were those complex wiggly interactions, they would not show up as hits from the GWAS studies since they're not additive, and they would not contribute to the large additive fraction of heritability but the other parts.