Biorecap: An R package for summarizing bioRxiv preprints with a local LLM

Automatic document summarization is very hard without knowing what purpose the summary serves. For what/whom is the package intended?

Evaluating the quality of summaries is also very difficult. How would you evaluate whether a bioRxiv paper is well summarized or not? In the scientific literature you will find that one way is to let human domain experts score various dimensions of the produced summaries e.g. using a 4-point Likert scale. Alternatively, professional summarizers produce "reference summaries" that serve as a model, and automatic summaries can be evaluated automatically against these using string overlap metrics.

I do not judge the package BUT have to note a very interesting thing: this package show a classical thing about human interaction with a computer, where data and code are mixed and user programming is a thing.

People could remain in a "development environment" not just to develop something or to manipulate numbers but for more generic data usage. Aside Quarto reports who can mix text, formatting and live code, like org-mode is another classical thing in a modern environment.

It's VERY positive to me, because it's a simple sign of another step toward classic computing, a thing we have lost and that's desperately needed.

Doesn't this mean that biologists have a problem writing effective abstracts in the first place? I thought the point of an abstract was the summarize and motivate.