They are trained to provide answers to users’ questions. If getting answers and haved their problem solved pleases the users, then one can say they are desgned to please users.
But no, they will not make things up just so a user will be pleased. To please someone is an emotional response. AI has no emotions.
I am talking about AI that are readily available to everyone. The example that I linked to earlier, which I will linked here again since no one goes back to read, used Google (which is Gemini).
It neatly provided 3 solutions to a Fusion sketch question. Click at the link to take a look and tell me in which way it tried to please the user other than giving out solutions.
Why would I train AI only on Reddit? Unless I want the AI tool to only provide answers based on Reddit posts, which doesn’t make sense. None of the readily available LLMs are trained only on Reddit.
The other points that you made about some answers are wrong on Reddit is actually one more reason that users should use AI as their first go-to venue. Imagine one gets a wrong suggestion from Reddit that is wrong and ruin one’s printer.
With the widespread access to AI tools, users no longer have to rely only on one person’s answer on Reddit, or this forum, or any other site alone.