The deeper issue this site highlights, is that failure modes for these kinds of complex automations (and LLMs are automations, not intelligence) are not only no

The deeper issue this site highlights, is that failure modes for these kinds of complex automations (and LLMs are automations, not intelligence) are not only non-linear, non-local - but unbounded and 'super-causal' (I'm trying to reach for some proper terminology to say that output y may have no relation at all to input x).