How web-enabled models cut hallucinations from 47% to 9.6% in AI program tests

https://atavi.com/share/y05cbuz1m69ve

The data suggests that adding controlled web access to a large language model can produce dramatic improvements in factual accuracy for enterprise use cases