I absolutely agree with you. This is not only limited to code unfortunately.
For most answers, Aria creates links to google searches on random words.
For example if Aria is composing a response about Tesla, instead of linking the word "Tesla" to the website's business, it links to a Google search for "Tesla". This could be useful in some cases, but most of the time it isn't.
I'm hoping the team will be looking at this improvement in the future.

I absolutely agree with you. This is not only limited to code unfortunately.
For most answers, Aria creates links to google searches on random words.
For example if Aria is composing a response about Tesla, instead of linking the word "Tesla" to the website's business, it links to a Google search for "Tesla". This could be useful in some cases, but most of the time it isn't.
I'm hoping the team will be looking at this improvement in the future.

@mohsenof this is not how large language models work I'm afraid. They are not google.

In addition to that, one of the most common warnings you read online about language models is that "it could provide the wrong information". Actually in the case of Aria there is a warning at the top of the Aria interface "AI responses can be inaccurate or misleading".
So the fact the language model made up a fake answer, or a different Mohsen, is not a defect.

If I search my name and surname, Aria gives me a list of facts about me, only one of them was correct. The other facts could be about other people in the world with the same name, or they could be invented.

For a more search-focused, and more factually accurate AI experience at the moment I would suggest you try Bing chat.