Right now all Transformer LLMs make up stuff, ie, hallucinate, because essentially they don’t really know what’s going on, but experiments like this should alert us as to what we’re enabling if and when we develop artificial neural networks that have reality understanding as our biological ones do.