Meta AI says Trump’s attempted assassination didn’t happen. Then, it blames hallucination

Readers help support MSpoweruser. We may get a commission if you buy through our links.

Read our disclosure page to find out how can you help MSPoweruser sustain the editorial team Read more

Meta was recently thrown under the bus for mishandling political content. The Menlo Park tech giant has previously slapped a fact-check label on a genuine photo of former President Trump after his attempted assassination, and its Meta AI has incorrectly claimed that the attempt did not occur.

Joel Kaplan, Meta’s VP for Global Policy, has issued an apology in a new blog post, attributing the company’s recent AI error—where the assistant incorrectly claimed that the attempted assassination of former President Donald Trump did not happen—to “hallucinations” in the AI’s responses.

He also says that the Meta AI — just like any other AI chatbot with knowledge cutouts like ChatGPT — has not always been the most accurate when it comes to breaking news due to reliance on outdated data. To avoid errors about such news, Meta initially blocked responses on the topic but still faced issues with inaccuracies.

Such a phenomenon is called “hallucinations” in the AI world. It’s an instance where a language model generates incorrect or misleading information that wasn’t present in its training data. That happens because the AI predicts responses based on patterns in the info it was trained on, rather than factual, real-time accuracy, so it had to yap and formulate answers.

Meta AI chatbot, which is available across Facebook, Instagram, and WhatsApp, is powered by the latest Llama 3.1 model with knowledge cutoff in December 2023. OpenAI’s ChatGPT, on the other hand, is in October 2023.

Rafly is a reporter with years of journalistic experience, ranging from technology, business, social, and culture. Currently reporting news on Microsoft-related products, tech, and AI on MSPowerUser. Got a tip? Send it to [email protected]