Why AI Invents Things: What Hallucinations Are and What Is Known About Their Origin
4 Articles
4 Articles
informacion.es
They once asked a chatbot about the doctoral thesis of Adam Tauman Kalai, one of the authors of an investigation into AI hallucinations. He responded with a convincing title. They repeated the consultation and offered another. Later, a third. None was correct. The case appears in an OpenAI investigation of hallucinations and summarizes a known frustration: the machine responds when it does not know.
A team of researchers from OpenAI and Georgia Tech asked a state-of-the-art open model about the birthday date of one of them, Adam Tauman Kalai, on one condition: answer only if he knew it. In three attempts he gave three different dates, the three false ones. The example opens the paper Why Language Models Hallucinate,
Coverage Details
Bias Distribution
- 100% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
