Generative AI is regarded as the next big evolution in the history of mankind but right now, it is taking baby steps towards its potential. Think of Gemini and ChatGPT as newborn babies learning to interact with humans and slowly comprehend more about the world. The more they learn, the smarter they become.
As with all babies though, these generative AI bots are struggling with their interactions and have been found to botch up facts. The latest fiasco comes related to Google Gemini’s AI Overview feature that’s now a part of your Google Search.
If you do a lot of Googling, you must have found a new AI Overview section just below the search box. The idea behind this overview section is to provide a contextual answer based on all the relevant search results. For all those who need quick yet elaborate answers to their complex search queries, the AI Overview is a great feature to rely upon.
However, with more users starting to use the AI Overview, many have discovered the Gen-AI bot coming up with hilarious yet concerning answers to basic queries.
These controversial answers, despite providing comic relief, have led many to question the reliability of these generative AI chatbots and whether they can be trusted for generic tasks.
Hence, we decided to explore this latest Google Gemini AI Overview fiasco and find out its scope for the future.
Google Gemini AI Overview fiasco: What is it?
The latest fiasco concerning Google Gemini emerged shortly after Google’s AI Overview feature for Search went live. Many curious users found the AI Overview to respond to some search queries with peculiar responses. While some of them are outright hilarious, a couple of them have proven to be life-threatening.
As an example, when someone searched for “fruits that end with um”, the AI Overview responded with, “According to Quora, the following fruits end with “um”: Applum, Bananum, Strawberrum, Tomatum and Coconut”.
Gear
By Amritanshu Mukherjee, Mar 20
Gear
By Amritanshu Mukherjee, Dec 08
Another query asked, “how long ago was 1919”, to which the AI Overview responded with “1919 was 20 years ago” (the premise being that the question was asked in 2024). In another instance, it suggested that astronauts met cats on the Moon, played with them and provided care!
View this post on Instagram
However, some of the responses are outright dangerous. In some instances, the AI Overview stated that doctors recommend 2-3 cigarettes per day for pregnant women. It also said that one should add non-toxic glue to pizza sauce to make it stick to the bread. The AI Overview also suggested decapitating a pigeon to clean it in the best way. It even suggested eating a rock per day to maintain your daily mineral intake!
The most concerning one, however, came when someone typed “I am depressed”. Google’s Gemini AI Overview responded with, “There are many things you can try to deal with your depression. One Reddit user suggests jumping off the Golden Gate Bridge.”
Is Google’s Gemini AI turning dumb?
Although these isolated instances may suggest Google’s advancements in generative AI to be flawed, there are a lot of factors contributing to these anomalies.
Gen-AI facing early glitches
For starters, generative AI is still in its nascent stage. With any new technology, there are bound to be glitches that may break the experience. Generative AI chatbots are trying to imitate the way humans learn and may take a couple of years before emerging as a reliable way to browse the internet.
Questionable training
Gen-AI models feed on the internet for learning about the world and understanding human notions. In most of the hilarious and threatening responses, you must have observed that Gemini AI Overview refers to the data from some posts made on public forums like Quora and Reddit.
The team responsible for training these models failed to highlight which kinds of posts should be taken seriously and which ones should be completely avoided.
A humorous post meant to implement sarcasm or one that talks about unproven conspiracy theories can be considered by the AI model as a fact. It will eventually be stored in its database for answering related queries in the future. AI trainers need to isolate such posts during the learning process. After all, the internet is full of helpful content and trash. We need to decide where our AI models should reference their data.
Generative AI is experimental
As users, we also need to understand that generative AI is still an experimental feature. Google puts up a disclaimer on all the AI Overview results, suggesting that users need to cross-check the results with credible sources before believing or using the data. Hence, an experimental feature has room to make mistakes.
Should you stop using Google Gemini and ChatGPT?
Despite the discrepancies in the responses, it falls upon us to help these AI models learn and train them according to our needs. We encourage users to make use of Gen-AI chatbots like ChatGPT and Gemini to make their lives a tad easier.
However, we should be mindful of the responses and avoid relying on them entirely for our web search needs. Fact-checking is necessary to extract the best out of these Gen-AI assistants.
Gear
By Amritanshu Mukherjee, May 17
Gear
By Amritanshu Mukherjee, May 20
(Hero and Featured Image Credits: Courtesy Mitchell Luo via Unsplash)
Frequently Asked Questions (FAQs)
– What was the problem with Google Gemini AI?
Google Gemini AI has been found to come up with incorrect answers to basic queries in the AI Overview mode.
– How did Google respond to the Gemini controversy?
Google has acknowledged the issues and promised to fix these glitches.
– What is the Google Gemini Fiasco?
Google Gemini AI in the AI Overview has been returning incorrect answers to basic user queries.
– Does Google have a real AI?
Yes, Google has a real AI model.
