A report on the benefits of AI was reportedly full of AI hallucinations

Trending 2 days ago

It was published by KPMG, 1 of nan world's 'Big Four' accounting firms.

A personification typing connected a laptop.

Junayed graphics/Shutterstock

In October past year, KPMG published a study titled Total Experience: Redefining Excellence successful nan Age of Agentic AI, which was astir really companies are utilizing AI to cater to customers' needs. KPMG is 1 of nan "Big Four" master services and accounting firms successful nan world, on pinch Deloitte, PricewaterhouseCoopers and Ernst & Young. Apparently, though, that study was afloat of AI hallucinations and included examples of agentic AIs that either did not beryllium aliases did not person nan capabilities KPMG stated successful nan paper. Investigators for GPTZero, nan shaper of an AI content discovery tool, recovered inaccuracies and clone footnotes each complete nan report, which were besides verified by nan Financial Times. 

In its study of nan investigation, GPTZero said that only 5 citations retired of 45 successful nan insubstantial accurately pointed to existent sources. A full of 28 citations paraphrased titles aliases added clone components to existent sources, while 12 were phrased excessively vaguely to find whether they really existed. GPTZero called nan creation of clone references by AI models "vibe citing."

In summation to nan clone aliases inaccurate citations, nan investigators besides recovered that astir half of nan claims successful nan insubstantial were clone aliases misattributed. They were "likely nan consequence of an AI investigation instrumentality over-complying pinch a petition to find examples of 'agentic AI' successful nan wild," GPTZero wrote. In 1 example, KPMG claimed that Emirates launched a mobile chatbot called Sara that tin talk to passengers and change their flights for them. Sara was a mobile adjunct launched successful 2023 and not an AI-powered chatbot, and it besides didn't person nan powerfulness to alteration bookings for passengers. 

KPMG besides claimed that Swiss multinational finance slope UBS integrated agentic AI crossed its "investment advisory, consequence guidance and compliance monitoring." The slope told nan Times that nan accusation was "factually incorrect." In different example, KMPG said that Swiss Federal Railways (SBB) has AI agents that tin thief passengers plan, book and optimize their trips based connected preferences, real-time conditions and c impact. An SBB spokesperson said that was "not accurate."

Papers by companies for illustration KPMG are typically cited successful different investigation papers and articles, since they're considered arsenic highly trusted sources. GPTZero main executive Edward Tian explained that error-riddled papers published by nan Big Four could "poison nan good of information" and could lead to second-hand AI hallucinations. A KPMG spokesperson told nan Times that nan institution "takes nan accuracy and integrity of its published contented seriously." KPMG has since pulled nan insubstantial and is now "reviewing nan circumstances surrounding its publication."

More
Source engadget.com
engadget.com