Google's AI Is Learning Internet Memes and Facing Issues

TapTechNews June 12th. Remember that incident where Google's AI search result told users to add glue to pizza? At that time, the foreign internet celebrity Katie Notopoulos actually made a 'glue pizza' and ate it, which caused a sensation online. But now there is a problem: Google's AI is learning these internet memes.

Admittedly, people seldom query such a question as 'how much glue to add to pizza', but considering the recent popularity of 'glue pizza', it's not entirely impossible. Security researcher Colin McMillen found that if you ask Google how much glue should be added to pizza, you won't get the correct answer - which is definitely not to add glue. Instead, Google will quote Katie's spoof article and suggest adding one-eighth cup of glue, which is obviously very dangerous. TheVerge verified this through searching.

The Verge said that this means that when netizens report to Google that its AI is wrong, it is actually 'training' it to continue to make mistakes.

Another problem is that due to the use of AI technology, Google now seems unable to answer questions about its own products. The Verge editor asked how to take a screenshot of the Chrome browser in incognito mode, and Google's AI gave two incorrect answers. One suggested taking a screenshot in a regular mode Chrome tab, and the other answer insisted that it is simply impossible to take a screenshot in Chrome incognito mode.

TapTechNews noted that Google CEO Sundar Pichai previously admitted in an interview that these 'AI summaries' features generate 'hallucinations' that are an 'inherent flaw' of large language models (LLM), which is the core technology of the 'AI summaries' features. Pichai said that this problem still has no solution so far (is still an unsolved problem).