Interesting #astromethods paper came out recently that looks at whether or not large language models can replicate astrophysics papers. The answer: No. 🔭☄️ They had a ~20% replication score, even with quite a lot of help. And this was just replicating papers - not doing anything new...