Hypothetically speaking, how feasible would it be to task an advanced model like GPT-6 Astra with designing a functional molecular assembler? Where would the main bottleneck be: safety alignment, lack of experimental data, or physical modeling? Curious to hear your thoughts!