Why Even Smart AIs Can Give Different Answers

You know how sometimes you ask two different people the same question, and they give you slightly different answers? Well, it turns out that even cutting-edge Artificial Intelligence (AI) models can do the same, even when they come from the same family. A recent study has highlighted that different versions of Anthropic's AI, Claude, behave in distinct ways when searching the internet.

The research looked at Claude and a specific version called 'Claude Code'. It found that when these AIs were tasked with searching the web, they didn't just find different information, they even chose to visit different types of websites. Claude Code, for example, seemed to prefer using Wikipedia more often. This suggests that even subtle differences in how an AI is designed or 'trained' can lead to very different outcomes when it's put to work.

For everyday Aussies, especially those running small businesses or relying on AI for research, this is a useful insight. It means you can't assume all AI models, even similar ones, will give you identical results. If you're using an AI to gather information, check which version you're using. You might get a different perspective or a different set of facts just by switching models or versions.

Think of it like this: if you ask for directions, one person might tell you the quickest route, while another might suggest the scenic route, even if they both know the city well. Different priorities lead to different answers. With AI, these 'priorities' are built into their programming and how they learn. So, while AI is incredibly powerful, it's a good reminder to be aware of which tool you're using and to perhaps cross-reference important information, just as you would with human sources.