Yesterday
After doing research on AI hallucinations and factual errors over the last week, the first time we accessed Google (GOOG) search this morning we were faced with this unusual answer to our question. The question was “When did the Wal-Mart (WMT)/Vizio (pvt) acquisition close?”, a simple enough question that we needed for another note. The answer was as follows, which was obviously incorrect.:
“The Walmart/Vizio acquisition closed on December 3, 2024. The deal, worth $2.3 billion, was finalized following the expiration of the federal regulatory waiting period. Walmart is now a wholly owned subsidiary of VIZIO, and VIZIO's Class A common stock is no longer listed for trading on the NYSE.”
Not only was this answer inaccurate, but the press release it referenced, which was issued by Wal-Mart, was not the source of the error, leaving the Google Search Overview LLM the responsible party. The primary model for Google Search Summaries is Gemini, although a number of LLMs are used in certain situations. Regardless of which LLM or combination made the error, it was an obvious one that the LLM created without a source or misunderstood the basic premise of the Wal-Mart press release. More likely the later as LLMs do not have contextual understanding and are primarily based on pattern matching which means answers must always be checked for accuracy.
“The Walmart/Vizio acquisition closed on December 3, 2024. The deal, worth $2.3 billion, was finalized following the expiration of the federal regulatory waiting period. Walmart is now a wholly owned subsidiary of VIZIO, and VIZIO's Class A common stock is no longer listed for trading on the NYSE.”
Not only was this answer inaccurate, but the press release it referenced, which was issued by Wal-Mart, was not the source of the error, leaving the Google Search Overview LLM the responsible party. The primary model for Google Search Summaries is Gemini, although a number of LLMs are used in certain situations. Regardless of which LLM or combination made the error, it was an obvious one that the LLM created without a source or misunderstood the basic premise of the Wal-Mart press release. More likely the later as LLMs do not have contextual understanding and are primarily based on pattern matching which means answers must always be checked for accuracy.
RSS Feed