Ideas:
“
Working on a series of connected papers on structural illusions generated by large language models (LLMs): the epistemic illusion (conflation of syntactic plausibility with truth), the phenomenal illusion (conflation of simulated affect with subjective experience), and the moral illusion (unwarranted attribution of normative competence). The argument is that these illusions are architectural properties of LLMs, rooted in the disjunction between stochastic core and abductive appearance, rather than correctable defects. Applications to clinical medicine and AI governance are under development.
”
Feedback