Mohit Iyyer#366
🤖 When 5 frontier LLMs are asked to write about if AI’s future has to be dystopian, they all converge on a broad, safe argument: let’s augment humans instead of automating work.
👩 Human writers take a wider range of sharper positions: AI is already used to micromanage workers; “better” AI requires worker protections and public institutions; decolonizing AI means changing who gets to shape and benefit from it.
We call this AI argument collapse: on debated topics, different LLMs converge on a small set of main arguments, supporting claims, and argumentative structures.
Newspaper op-eds, position papers, even tweets… many forms of public debate risk being flattened if we are not careful about AI-assisted / AI-authored pieces. At scale, many “reasonable” AI-written arguments may become “reasonable” in the same exact way.
Read more in our thread 👇
From op-eds in newspapers to NeurIPS position papers, AI is increasingly shaping long-form public discourse. Its arguments seem plausible, but beneath surface fluency, we find argument collapse: different LLMs converge to the same main & supporting arguments and structure.