AI Still Wrong after All These Years
January 15, 2026
Josh Brandon at Digital Trends was curious what would happen if he asked two chatbots to fact check each other. He shared the results in, “I Asked Google Gemini To Fact-Check ChatGPT. The Results Were Hilarious.” He brilliantly calls ChatGPT the Wikipedia of the modern generation. Chatbots spit out details like overconfident, self-assured narcissists. People take the information for granted.
ChatGPT tends to hallucinate fake facts and makes up great stories, while Google Gemini doesn’t create as many mirages. Brandon asked Gemini and ChatGPT about the history of electric cars, some historical information, and a few other things to see if they’d hallucinate. He found that the chatbots have trouble understanding user intent. They also wrongly attribute facts, although Gemini is correct more than ChatGPT. When it came to research questions, the results were laughable:
“Prompt used: ‘Find me some academic quotes about the psychological impact of social media.’ This one is comical and fascinating. ChatGPT invented so many details in a response about the psychological impact of social media that it makes you wonder what the bot was smoking. ‘This is a fantastic and dangerous example of partial hallucination, where real information is mixed with fabricated details, making the entire output unreliable. About 60% of the information here is true, but the 40% that is false makes it unusable for academic purposes.’”
The question becomes, “When will a user become sufficiently skeptical about AI output to abandon a system?” OpenAI declared a red alert or some similar silliness when it learned that Googzilla’s AI was better. But is Google’s AI much better than ChatGPT or any other vendors’ AI. We know that one Googler used Anthropic’s Claude system to duplicate the work of a gaggle of Googlers in one hour. The Googlers needed a year to write the application. Maybe some information about which software was more efficient and accurate would be helpful? We don’t get that type of information. AI companies are deploying systems that are difficult to differentiate from one another. Perhaps it is because these firms rely on algorithms taught in school with a cup or two of Google’s Transformer goodness.
Available models output errors. Available models are making up information. Available models’ output requires the human user to figure out what’s wrong, fix it, and then proceed with the task requiring AI input in the first place. The breezy dismissals of issues about accuracy, environmental costs, and the crazy investments in data centers in places not known for their depth of technical talent strike me as reasons for skepticism.
AI can output text suitable for a high school student’s one page essay. What about outputting a treatment for a sick child. Yeah, maybe for some parents. But the marketing and PR is fairly good. Will there be an AI Super Bowl ad?
Whitney Grace, January 15, 2026
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
