1/n Can LLMs perform scientific research? And, can they do so while enhancing key scientific values including transparency, traceability & verifiability? Check out our preprint on the "data-to-paper" platform. w the amazing Tal Ifargan Lukas Hafner

2/n Imitating human science, data-to-paper guides interacting LLM and rule-based agents through a multistep process: research goal, hypothesis, literature search, research plan, coding and debugging, interpret results and write a complete *human-verifiable* research paper.

3/n Since our sneak peek of data-to-paper in June we have added several new aspects to assure and enhance traceability and verifiability. IMO now even beyond the standard of human-driven research. In particular:

4/n “Data-Chaining": by tracing information flow through the steps, data-to-paper creates “data-chained” manuscripts, where results, methodology and data are *programmatically* linked. See video above, or try out "click-tracing" results in this example ms:

5/n Co-pilot: A human-AI co-pilot app allows users to oversee and direct the process by providing human reviews at each research step.

6/n Not (yet?) as novel as high-end research but does reach de novo results in different fields. We tested it in epidemiology, social networks, ML modeling & clinical datasets. Can autonomously tackle simple research questions. Human copiloting becomes critical for complex goals.

7/n Repo is now open. We invite people to try it out. Currently most suitable for fairly simple datasets with simple research goals, where we want to raise and test a statistical hypothesis. Help in developing and extending is also very much welcome.

8/n LLMs are here to stay and are already used extensively in science doing (sadly sometimes undisclosed Our approach is designed to provide ways to use LLMs powerfully, yet also transparently and responsibly.

9/9 LLMs have a huge potential for accelerating science. We hope our project stimulates discussions on how to harness AI for science, while fostering, not jeopardizing, accountability, transparency and verifiability.

@TalIfargan @LostInTranscrip Dark mode for this paper for those who read at night 🌙