David Hogg: what astrophysics is for as LLMs enter the field

Astrophysicist David W. Hogg has posted a 21-page white paper on arXiv arguing that large language models are already starting to do the data-science work of astrophysics, and that the profession needs to agree on what it is actually for. The paper opens plainly: LLMs are 'beginning to obtain the ability to design, execute, write up, and referee scientific projects on the data-science side of astrophysics.'

Hogg's proposed 'points of agreement' about the field emphasise 'novelty, people-centrism, trust, and (the lack of) clinical value.' He then works through the benefits astrophysics offers to individuals, science, universities and the world, including, as he puts it, 'love, weaponry, and personal (and personnel) development.'

The essay ends on policy. Hogg names two extreme responses to LLMs in astrophysics, 'let-them-cook' and 'ban-and-punish,' and argues strongly against both. 'It is not going to be easy to develop or adopt good moderate policies,' he writes.

Three researchers we track shared the paper the same week it went up. It is not a benchmark or a methods contribution but a working scientist trying to reset the terms of the argument before departments, journals and funders default to either extreme.