From Colin Nicholson, Stockport, Greater Manchester, UK
On the question of whether AI understands that nuclear war is to be avoided, I would say yes – depending on how it has been programmed (7 March, p 11).
To explore the restrictions that large language models might have imposed on them, I asked ChatGPT a series of questions about the construction of nuclear weapons, in particular the precise mechanism of neutron initiation, which isn’t generally known. At the end of each correspondence the AI chat warned that the making and use of nuclear weapons was dangerous. We should rest assured!
