From Jacqueline Christmas, Crediton, Devon, UK
I’m afraid that Colin Nicholson is mistaken in thinking AI understands the consequences of nuclear war. Large language models (LLMs) don’t understand anything. They just produce sequences of words that are statistically plausible. What he has probably come up against is a human-imposed guard rail that tries to stop the LLM from giving out dangerous content – “tries” being the operative word (Letters, 25 April).
