PLOS-LLM: Can and should AI enable a new paradigm of scientific knowledge sharing?

Most people agree that the traditional academic publishing model is not working 1,2.Even editors of prestigious journals have been saying this for some time.There is, however, less consensus about which of the many problems are most pressing.Is it paywalls limiting access, especially for those outside of rich countries/institutions? Or is the inefficiency inherent in reformatting papers as they bounce between dated online submission systems with login systems from the last century?Perhaps the key problem is the way that journal editors, with variable urgency, decide what to publish in their prestigious journals, juggling commercial pressures to attract attention/clicks/advertising with attempts to identify what research is most important/ transformational?Add in that the market is dominated by a tiny number of publishers-in 2013 for example, over half of all published papers were published by just five publishers 3and it is perhaps unsurprising that it's getting harder and harder to attract peer reviewers 4, further slowing the system down.And this is without even getting into the predatory business models of the less scrupulous journals 5.What is perhaps even more concerning is how these problems are likely to influence not just the dissemination of research, but the whole system of scientific inquiry from decisions about what 'publishable' work to focus on to which results to and do not get into the public domain.The ongoing shift towards open access publication represents a set of important, yet far from inclusive 6, changes in the academic publishing landscape as do the rise in use of preprints, emerging models of transparent, progressive, 'post-publication' peer-review.Social media is also radically changing how, and what, science is shared.But is it time for a much more radical disruption to the ecology of scientific knowledge?In this article, having touched on some of the many current problems with academic publishing, we propose a new model for scientific knowledge sharing.We believe that emerging technologies, specifically foundational large-language models 7 trained with human supervision and supported by semantic search, may enable a radical re-thinking of what the 'journal of the future' might be.We hope to provoke discussion and debate about how such a revolution could, and should, be more open, transparent and efficient; something that-unlike the status quo-is fit for the 21 st century, benefiting both researchers and all of society.It is worth beginning by remembering that-although journals do sometimes feel like they have been around forever-there was a time when a different model dominated.Scientific journals as we known them were introduced in 1665 with the publication of the journals including the Journal des Sc ¸avans in France and the Philosophical Transactions of the Royal Society of London in England.They were founded to advance scientific knowledge by building on