There are 4 base types (A,T,C,G) so it's only 2-bit values. There is also some redundancy coding in the amino acid sequence that follow the DNA. So if you compu

There are 4 base types (A,T,C,G) so it's only 2-bit values. There is also some redundancy coding in the amino acid sequence that follow the DNA. So if you compute the information of the codons sequence it's even lower. It's quite compact. What would be interesting to know, is the percentage of programs still running when a few bits are modified, to know how sharp is the search space. Exploring living programs space, you probably can use some advance neural network techniques, like GPT-2 to explore it more efficiently, directly in its compressed representation to provide an high level interface with slider controlling the high-level properties.