OP here: We have data source powered research papers in the making, that reports on experiments executed with codegen - really hopeful that the papers can be informative at a minimum.
It's clear that superhuman citation depth & breadth is already imminent. Hopefully we can push hallucinations to near-zero with the next generation of models.
