AI and the future of human expertise: math as the test case

A thread titled 'Why do we need human mathematicians anymore?' surfaced a broader anxiety that keeps returning to HN in different forms: if AI can do the intellectual work, what is the human role. The math framing is specific and useful because mathematics is one of the domains where AI progress has been most verifiable and most surprising. The original article, by Po-Shen Loh, wrestles with this directly rather than dismissing it.

Commenters pushed back in two ways. Some argued that the same logic applies to every profession and the conclusion that we won't have jobs is wrong in the short run and uncertain in the long run. Others pointed to the LLMentalist Effect thread, which argues AI looks more capable than it is because it succeeds at the visible, easily evaluated tasks while failing invisibly on harder ones. That thread was itself accused of being AI-written, which is either ironic or evidence for its thesis, depending on your priors.

The Heretic thread, covering a tool that removes restrictions from language models through automated abliteration pipelines, adds another dimension: the question isn't just what models can do, but what they are allowed to do and who controls that.