There are fun issues about LLMs like the one about if it is better to go to a car wash on foot or by car if it is just 50 meters away. LLMs (as of now) will confidently tell you that of course going on foot is better in that case.
I am sure that the companies and developers will fix this eventually. But it will not solve the problem. It will be just another layer of camouflage. The AI will still not really understand why the answer is bullshit. It will just predict best words better next time.
By hiding such systematic failures by overfitting responses so people do not make fun of the systems will just hide the problem deeper and make it more unpredictable.
Unless we figure out how to really make the system understand context and relations, it is always just another layer of bullshit on top of bullshit and you never know when it will blow in your face. And in more subtle and unpredictable way than a simple question about going to a carwash.