Poisoning LLM? I remember at least one study about it. Perhaps but it would have to be very subtle otherwise they might flag your content and shadowban it from future training.

My personal opinion is that trying to make LLMs inefficient is futile and causes even more waste of earth resources.

It's like trying to stop a coal-powered locomotive by parking a truck on the rails. You're just going to make the thing burn more coal before reaching it's destination.

I wasn't talking about poisoning LLM for generally degrading performance/improvement. But for manipulating answers for propaganda/ideological reasons.