Breakdown
- Google's new AI Overview snippets have been telling users to eat glue and chug pee
- The AI is seemingly using some information from Reddit comments from years ago, after Reddit's deal with Google a few months ago
- Google has introduced a 'Web' filter for web link-only results, and a user-created extension also kills the AI Overviews
Most people prefer simple web results when searching on Google, but the company has different ideas about what the general audience needs. Google is gradually rolling out Google AI Overviews to users, providing snippets that use information to give quick answers, and the results are as impressive as you'd expect.
Most companies are either integrating useful AI tools into your favorite devices, such as the Galaxy AI features in your Samsung smartphone, or creating the best AI chatbots like ChatGPT 4o, Copilot, and more. However, Google is eager to introduce AI elements into your daily search habits, and it hasn't had a great start.
A range of posts on social media showcases the terrible suggestions that the AI Overview is providing. One example appears if you search 'cheese not sticking to pizza,' and while some information is useful, one bullet point suggests users should put non-toxic glue on their pizza to keep the cheese stuck.
While that's downright silly, some results are more harmful. Searching 'how to pass kidney stones quickly' will see the AI suggest helpful things like drinking plenty of fluids before recommending drinking 2 quarts or liters of urine. At least Google recommends it be light in color.
There are obvious concerns about this feature causing more harm than good, but amusingly enough, some of the information comes from Reddit. Many people had conspiracy theories that Google's favoritism of Reddit was for sinister reasons, and that theory only heightened once Google and Reddit signed a deal for the latter to provide data for Google's AI. But, the "adding glue to your pizza" suggestion is a Reddit comment from 11 years ago.
Even worse, searching 'what is in Google's AI dataset' will give you a list of datasets that Google's AI uses for information. Right at the bottom, underneath the list, it states that "Google's AI models have been trained on child sexual abuse material (CSAM)," which is shocking, to say the least. However, we wouldn't be surprised if that's for child safety rather than anything nefarious.
Google clearly understands the disappointment with the AI Overview feature. Google is allowing users to filter 'web' results only, removing AI features and shopping links, while some users are actively installing a Google AI-blocking extension for Chrome.
It's safe to say that Google's entry into the AI market isn't going too well. To be honest, it's not going well for many AI-focused companies right now, with OpenAI facing a lawsuit due to ChatGPT 4.0's voice sounding like Scarlett Johansson. It's not a great time to be in the AI industry, it seems.