The instructions to code for general intelligence must be small enough to fit in human DNA alongside the instructions for a living body. So very, very small in computer terms. The algorithm could be encrypted into a PNG image and couldn't even be censored in today's China.

That's very much an apples to oranges comparison and that quantification makes no sense. DNA is the blueprint for a biological hardware on which, in the right conditions, a general intelligence can self-organize. It's not the code for the intelligence itself. And there's a whole slew of epigenetic effects in human development that affect intelligence too, so DNA is far from the entire equation.

But even ignoring that, it's not small by any means. If you do a crude conversion of 1 DNA base pair = 1 line of code (which would be the closest approximation, as each is essentially a single instruction step in their respective environments), the human genome is 6.4 billion base pairs long (3.2 if you ignore chromosome duplication).

So at best you can say, a general intelligence AI should be possible with <3.2 billion lines of code. (That is base code, not training data) That is not "small in computer terms". The entire codebase of Google is estimated to be the biggest in the world at 2 billion lines now. And that's not accounting for how code efficient an AI would need to be to match a biological system with millions of years of optimization behind it.

The instructions to code for general intelligence must be small enough to fit in human DNA alongside the instructions for a living body. So very, very small in computer terms. The algorithm could be encrypted into a PNG image and couldn't even be censored in today's China.

That's very much an apples to oranges comparison and that quantification makes no sense. DNA is the blueprint for a biological hardware on which, in the right conditions, a general intelligence can self-organize. It's not the code for the intelligence itself. And there's a whole slew of epigenetic effects in human development that affect intelligence too, so DNA is far from the entire equation.

But even ignoring that, it's not small by any means. If you do a crude conversion of 1 DNA base pair = 1 line of code (which would be the closest approximation, as each is essentially a single instruction step in their respective environments), the human genome is 6.4 billion base pairs long (3.2 if you ignore chromosome duplication).

So at best you can say, a general intelligence AI should be possible with <3.2 billion lines of code. (That is base code, not training data) That is not "small in computer terms", it's unheard of and practically impossible to run effectively on current hardware if you're not willing to wait a decade for the output to the question "describe a flower"