I thought I’d have a play with AI today because I keep on hearing that it’s getting so much better. I asked it who Blackpool FC’s best manager was. OK, that’s subjective but it can still put the facts together and analyse them to come to a conclusion.

The AI answer - Malcolm Allison.

Now I knew the name but thought that he must have been a manager before my time so I asked “When did Malcolm Allison manage Blackpool FC?”

The AI answer - 1959-1965. He left to manage Preston North End and took Blackpool to the FA Cup Final in 1961.

A quick Google found…

Malcolm Allison never managed Blackpool, or PNE. He was still a player in 1959 and only started to manage teams from 1963. The 1961 FA Cup Final was Spurs vs Leicester.

Now I’ll agree that Ollama with the Qwen3.5 LLM probably isn’t the most bleeding edge environment to use for such questions, but it still leaves the question of why AI likes to lie so much, or perhaps to be kind - hallucinate so much? Interestingly the long screed of explanation for both questions I asked included a check for hallucinations - which clearly wasn’t good enough!

For some reason AI always insists on giving an answer, even if it’s completely wrong and made up. I think it will be a much better tool if/when it gains the ability to say “I don’t know”.