Patrick BiddiscombeNew Breed · Founder & CEO of New Breed | HubSpot-Led Growth | Advisor to HubSpot & FinQore

Why does Ai hallucinate?

Name: Why does Ai hallucinate?
Uploaded: 2024-05-31T13:42:36+00:00
Duration: 37 s
Channel: Patrick Biddiscombe
Description: A short YouTube commentary (Patrick Biddiscombe, May 31, 2024) about why AI systems hallucinate. The clip quotes Sam Altman saying GPT-5’s focus is on…

31 May 2024 · 1 min

Watch on YouTube

0:37

A short YouTube commentary (Patrick Biddiscombe, May 31, 2024) about why AI systems hallucinate. The clip quotes Sam Altman saying GPT-5’s focus is on accuracy, and highlights that model outputs vary with repeated prompts (e.g., asking OpenAI 10,000 times yields different answers). The speaker frames hallucination as an inherent creativity feature of current AIs that must be countered when higher accuracy is required.

Chapters

Sam Altman in an interview recently, they asked him, what are you excited for in GPT5? He literally said accuracy, right? If you ask OpenAI the same thing 10,000 times, you will never really get the same answer, or you will get a variety of different answers. And the problem Google have is what AI tends to do. You know, it's sometimes it gives you the great answer and you have a really good experience. Other times it gives you an incorrect answer, it completely hallucinates. Hallucination is a feature of AI. When it doesn't know the answer, it creatively tries to give you the answer. Creativity is a feature of AI. It is not not a bug, but it is something that companies, when they need to increase the accuracy, have to try to counteract.