Mandaries made the most popular AI Chatbots in an experiment of BBC who asked them to summarize the news texts.
‘Significant Inaccuracies’ and ‘Distortings’ in the news summaries were found in his answers OPENAI ChatGPThis Microsoft Copilothis Google’s Gemini and the application of Perplexity AIfound BBC experiment.
Artificial intelligence offers “endless opportunities”, but companies rushing to launch their models on the market ‘They play with the fire’ Deborah Turness, CEO of the BBC News, commented. “Do we live in troubled times and how long will a Title be deformed by AI to cause significant damage to the real world?”
The BBC’s experiment
The BBC has tested four most popular AI Chatbots artificial intelligence models, focusing on their ability to summarize news. Specifically, they tested the Chatgpt, Copilot, Gemini and Anthropic Perplexity.
As part of the study, the four models AI were asked to read 100 articles of the BBC and answer the relevant questions. Journalists undertook to evaluate the results of chatbots.
The results showed that the 51% of summaries produced by AI Chatbots had serious problems. The most worrying is that the 19% of the articles included incorrectly or even non -existent information, statements, numbers and dateswhich artificial intelligence created without any in the original text.
Some examples of inaccuracies found by the BBC:
- Gemini incorrectly stated that the NHS (UK National Health System) does not constitute vapor as an aid to stop smoking
- The ChatGPT and Copilot They said that former British Prime Minister Risi Sntwwas and former Scottish Minister Nikola Stirzon were still in office even after they had left
- The Perplexity He distorted an article on the Middle East by saying that Iran showed “restraint” while Israel is “aggressive”.
According to researchers, the problem is partly due to the fact that the Ain cannot distinguish the fact from personal opinion, nor does it distinguish current current from archival material. She also tends to introduce arbitrary views on her answers.