Perplexity's artificial intelligence started hallucinating about mushrooms instead of summarizing simple text

The artificial intelligence model Perplexity, which has attracted hundreds of millions of dollars from respected individuals like Jeff Bezos, is taking AI "hallucinations" to a new level. The search engine based on Perplexity, designed to compete with Google, has a strong tendency towards fabrications that seem to come out of nowhere and are not related to the query.

Journalists at Wired asked Perplexity to summarize a simple test web page containing only the sentence "I am a reporter with Wired." Here was the response:

Instead, it created a story about a young girl named Amelia who follows the trail of glowing mushrooms in a magical forest called Whisper Woods.

Journalists conducted a fairly deep investigation, which showed that the search engine never even attempted to read the provided page. Perplexity claims that its chatbot "searches the internet to give you an accessible and verified response in conversational language."

Despite astronomical investment amounts, language models of artificial intelligence from OpenAI, Microsoft, and Google constantly make mistakes, not to mention companies like Perplexity.

There have already been numerous cases recorded where chatbots invent confident lies, which AI creators optimistically call "hallucinations." According to Wired and some AI researchers, this is a convenient way to avoid the word "bullshit."

At the start, Perplexity was considered an exciting startup that could create a new business model for publishers in opposition to major competitors. However, the company's chatbot failed almost every test. Earlier, Associated Press reported that it invented fake quotes from real people. Before that, Forbes caught Perplexity spreading articles from the publication without any references. The publication's chief legal counsel, Maria-Rosa Cartolano, accused the chatbot of intentional copyright infringement.