On the birdsite I read a thread by someone using ChatGPT with questions about a sensitive political topic. It reminded me of something I realized years ago about neural nets: they can't explain the reasons behind their output aside from claims that statistically they are right x% of the time on data they've been tested with so far. That doesn't work in the courts or in any context where you have to cite sources and argue about which to rely on when they conflict.
#ChatGPT #AIreliability