Why does AI hallucinate?<\/h2>\n\nAI models are trained on large datasets to make predictions, but they can sometimes “hallucinate,” meaning they produce false or inaccurate statements. This often happens due to incomplete or biased training data.<\/p>\n
Jesse Kommandeur, a strategic analyst at the Hague Centre for Strategic Studies, likens it to baking a cake without a full recipe\u2014where you rely on past experiences to make educated guesses. Sometimes the result is good, but other times it fails.<\/p>\n
\u201cThe AI is essentially trying to \u2018bake\u2019 the final output (like a text or decision) using the incomplete \u2018recipes\u2019 it has learned,\u201d Kommandeur explained in an email.<\/p>\n
There have been several high-profile instances where AI chatbots have given false or misleading answers. For example, some lawyers submitted fake legal cases after using AI models, and earlier this year, Google’s AI<\/a> summaries were found to provide inaccurate information.<\/p>\nA 2023 analysis by the company Vectara revealed that AI models hallucinated between 3% and 27% of the time, depending on the tool. Additionally, the non-profit Democracy Reporting International warned ahead of the European elections that none of the most popular chatbots delivered \u201creliably trustworthy\u201d answers to election-related questions Microsoft’s AI.<\/p>\n
Could this new tool address the issue of hallucinations?<\/h2>\n
\u201cGenerative AI doesn\u2019t truly reflect, plan, or think; it simply responds sequentially to inputs. We\u2019ve seen the limitations of this approach,\u201d said Vasant Dhar, a professor at New York University\u2019s Stern School of Business and Center for Data Science.<\/p>\n
\u201cWhile it\u2019s possible that the new correction capability will reduce hallucinations, it\u2019s virtually impossible to eliminate them completely with the current architecture,\u201d he noted. Ideally, Dhar said, a company would aim to claim a reduction in a specific percentage of hallucinations.<\/p>\n
\u201cThat would necessitate a substantial amount of data on known hallucinations, along with testing to determine if this prompt engineering method effectively reduces them Microsoft’s AI. That\u2019s quite a demanding task, which is why they haven\u2019t made any quantitative claims about how much it reduces hallucinations,\u201d he said.<\/p>\n
Kommandeur reviewed a paper that Microsoft confirmed was published regarding the correction feature. He noted that while it \u201cappears promising and employs a methodology I haven\u2019t seen before, it\u2019s likely that the technology is still evolving and may have its limitations Microsoft’s AI.\u201d<\/p>\n
Gradual Enhancements<\/h2>\n
AI models are trained on large datasets to make predictions, but they can sometimes “hallucinate,” meaning they produce false or inaccurate statements. This often happens due to incomplete or biased training data.<\/p>\n
Jesse Kommandeur, a strategic analyst at the Hague Centre for Strategic Studies, likens it to baking a cake without a full recipe\u2014where you rely on past experiences to make educated guesses. Sometimes the result is good, but other times it fails.<\/p>\n
\u201cThe AI is essentially trying to \u2018bake\u2019 the final output (like a text or decision) using the incomplete \u2018recipes\u2019 it has learned,\u201d Kommandeur explained in an email.<\/p>\n
There have been several high-profile instances where AI chatbots have given false or misleading answers. For example, some lawyers submitted fake legal cases after using AI models, and earlier this year, Google’s AI<\/a> summaries were found to provide inaccurate information.<\/p>\n A 2023 analysis by the company Vectara revealed that AI models hallucinated between 3% and 27% of the time, depending on the tool. Additionally, the non-profit Democracy Reporting International warned ahead of the European elections that none of the most popular chatbots delivered \u201creliably trustworthy\u201d answers to election-related questions Microsoft’s AI.<\/p>\n \u201cGenerative AI doesn\u2019t truly reflect, plan, or think; it simply responds sequentially to inputs. We\u2019ve seen the limitations of this approach,\u201d said Vasant Dhar, a professor at New York University\u2019s Stern School of Business and Center for Data Science.<\/p>\n \u201cWhile it\u2019s possible that the new correction capability will reduce hallucinations, it\u2019s virtually impossible to eliminate them completely with the current architecture,\u201d he noted. Ideally, Dhar said, a company would aim to claim a reduction in a specific percentage of hallucinations.<\/p>\n \u201cThat would necessitate a substantial amount of data on known hallucinations, along with testing to determine if this prompt engineering method effectively reduces them Microsoft’s AI. That\u2019s quite a demanding task, which is why they haven\u2019t made any quantitative claims about how much it reduces hallucinations,\u201d he said.<\/p>\n Kommandeur reviewed a paper that Microsoft confirmed was published regarding the correction feature. He noted that while it \u201cappears promising and employs a methodology I haven\u2019t seen before, it\u2019s likely that the technology is still evolving and may have its limitations Microsoft’s AI.\u201d<\/p>\nCould this new tool address the issue of hallucinations?<\/h2>\n
Gradual Enhancements<\/h2>\n