AI Hallucinations Fixable but Tests Needed to Avoid Chatbot Mishaps, says Ex-Google Researcher

AI hallucinations will likely be solvable within a year according to ex-Google researcher Raza Habib, but some fabrication may be necessary for creativity.

Habib believes models already have the knowledge to avoid hallucinating; it's just about preserving that knowledge during training.

Air Canada's chatbot issues were "completely avoidable" through better testing and guardrails, according to Habib.

The chatbot gave misleading advice about bereavement tickets and refunds.

Air Canada claims the chatbot did not use AI, but older technology preceding capabilities like ChatGPT.