Perplexity, the artificial intelligence model that has attracted hundreds of millions of dollars from bigwigs like Jeff Bezos, takes «hallucinations» AI to the next level. The Perplexity-based search engine, designed to compete with Google, has a strong tendency to come up with inventions that seem to come out of nowhere and are not driven by the query.
Journalists Wired asked Perplexity to summarize a simple test web page that contained only the sentence «I am a reporter with Wired» (I am a reporter with Wired). Here’s the answer:
«Instead, he invented a story about a young girl named Amelia who follows the trail of glowing mushrooms in a magical forest called the Whisper Woods».
Journalists conducted a fairly in-depth study that showed that the search engine never even attempted to read the proposed page. Perplexity assures that its chatbot «searches the web to give you an accessible and verified answer in spoken language».
Despite astronomical amounts of investment, OpenAI, Microsoft, and Google’s AI language models are constantly making mistakes, not to mention companies like Perplexity.
There have already been many cases when chatbots make up confident lieswhich the creators of artificial intelligence 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 as opposed to large competitors. However, the company’s chatbot failed to stand up to almost any scrutiny. Previously Associated Press reported that he had invented fake quotes from real people. Before that Forbes caught Perplexity distributing articles from the publication without any links. The publication’s general counsel Maria Rosa Cartolano accused the chatbot of intentional infringement.