As models continue to evolve it may not even need to cheat. Since web scale data is already part of pre-training this info is in principle available for most b

As models continue to evolve it may not even need to cheat. Since web scale data is already part of pre-training this info is in principle available for most businesses without a web search. The exceptions would be if it’s recently added, or doesn’t appear often enough to generate a significant signal during training, as in this case with a really small business. It’s not hard to imagine base model knowledge improving to the point where it’s still performing at almost the same level without any web search needed.