So do you remember the crazy answers Google’s AI Overview was giving. Now, Google is not gonna let its users the curiosity unexplained. It comes up with an explanation at this blogpost.

The blog first gives an overview of the “Overview”. Unlike the other chatbots and LLMs (yeah, you know which ones they are referring to), AI Overviews are not just generating response based on training data. It takes into consideration web ranking systems so as to provide users with a reliable experience with relevant links. In their words, they don’t just hallucinate.

Where can it go wrong, then? Well, if it misinterprets queries or the language nuances on the web, or if there is not enough reliable information available.

So, what happened? Well, some of the screenshots, according to the blog, are fake. There were never such responses to any queries that said to smoke while pregnant. AI Overviews did give out odd answers to many nonsensical and satirical queries. Like, who would want to ask how many rocks to eat in a day? The answer that the AI picked up is from a satirical content that was republished on a geological software provider’s website. If you go there, you could see a note from the publishers to the AI visitors. And the other set of queries that gave incorrect answers like putting glue to pizza, well it came from forums that are sometimes authentic and informative, but sometimes, wrong and satirical.

Yeah, they did come with improvements. 

They made better detection mechanisms to identify nonsensical questions. Limited the use of user-generated contents that could give misleading suggestions. For hard news (news that include politics, business, international affairs which are usually timely and consequential), they have stopped showing AI Overview. They have added triggering refinements to health-related topics.

With modesty, they conclude the blog, having agreed to keep learning and improving combined with the 25 years of experience. This statement of experience is so relevant. This is something others lack. This makes the Google a great competitor plus leader in the race of establishing a powerful AI. In the interview, Geoffrey Everest did mention that Google had been ready with AI but were concerned at the beginning. Later, when OpenAI released theirs, Google had to soon follow up to be not left behind.