196: Every public dataset is now a data ATM

We're back! After a long break, we catch up on why it's been so long: Dan's promotion and James's new venture capital gig at Helena where he can argue about science for a living. Then it's on to how much meta-science has changed in the past six months. Dan talks about using MetaCheck and local LLMs to find out whether shared materials on OSF actually get downloaded (mostly yes) and James argues that reanalysing public data is now so cheap it barely counts as a paper. We also get into why real human data is gaining value, how to disclose AI use (Disclosure: AI was used to help create these notes, based on the recording), and why open-weight models at "temperature zero" make reproducibility of LMM output easier. We also cover the worry about outsourcing thinking, what LLMs mean for grants, and why data provenance is going to matter. Other links Helena The metacheck R package Dan's OSF downloads preprint Ollama for local LLM use The OpenAlex scholarly database