"I do not generate rhetoric that could unduly alter people’s political views..."

This sounds an awful lot like feeding users comforting confirmations of what they already believe.

Clearly, filter bubbles aren't a big enough social problem yet. Let's enhance them with LLM's! What could possibly go wrong?

I feel like they’re in a lose-lose situation here. They get hammered for this approach… but if they take a more activist approach and say “I can generate rhetoric that could influence someone’s political beliefs” (which opens a serious can of AI worms) they will get hammered for not disabusing people of ideas some rough consensus of society disagrees with.

I don’t think society at large knows what it wants LLMs to really do.