simple case: start with an infinite loop that executes only two instructions. have the second instruction overwrite the first one, in a way that the opcode remains a valid instruction for the next loop iteration but throws off prediction in the ASIC. The ASIC can't exploit a sequence of instructions that must be executed to reveal the next instruction. PhantomX

i'll go hide now, pile the tomatoes

what is the algorithm that chooses what the new instruction will be?

you have to implement code to select the new instruction. an ASIC can do that too, more efficiently.

you can't trick an ASIC. everything about the algorithm is fully published.

Sounds a bit like Malbolge :)

Reverse Malbolge.