There's more low-hanging fruit in interdisciplinary work thanks to LLMs
post by ChristianKl · 2025-05-07T19:48:37.250Z · LW · GW · 2 commentsContents
2 comments
I'm right now doing conceptual theoretical work about how the human fascia system works. While I do rely on some original conceptual insights that I have e come up with on my own, Gemini 2.5 Pro massively speeds up my conceptual work. Being able to formulate a principle or analogy and then having Gemini apply it, is very useful.
There are a bunch of scientific fields where we currently have a lot of experimental data but lack coherent theory to interlink the experimental findings. Based on my own experience, current LLMs seem already to be powerful enough to help bridge that theory gap. Being able to ask "Hey, does field XYZ have any useful insights to the problem I'm tackling?" is also very helpful for making progress in theory.
The LLMs also solve a key problems that autodidact have when it comes with existing scientific fields. If you have a new idea, they are good at telling you the main criticisms that would come from an orthodox researcher in a field. We might see a rise in interdisciplinary work that didn't happen in the past because of academia's hyperspecialization.
People frequently say that progress in science has stalled because there's little low-hanging fruit. When it comes to doing certain interdisciplinary work, it's now a lot easier to pick the fruit. If you are right now starting a scientific career, think about what kind of interdisciplinary work you might do, where it's now easier to make progress because of the existence of LLMs.
If you have a research question, one approach you can do is to ask a reasoning model to create a debate between two highly skilled researchers with different approaches to debate your research question. You might learn valuable insights about your research question this way. Besides taking existing researchers in the field, asking the LLM to simulate philosophers and tell the LLM that the philosophers understand all the facts about a field, might give you valuable insights of how insights that philosophers found through a lot of hard work translate into individual fields.
It's not clear what the best approaches are to get the LLM to help you with interdisciplinary work, but there's a lot of fruit out there to be picked right now.
2 comments
Comments sorted by top scores.
comment by rain8dome9 · 2025-05-10T06:27:19.689Z · LW(p) · GW(p)
Please give an example of the output that the LLM produces that is so useful.
Replies from: ChristianKl↑ comment by ChristianKl · 2025-05-10T19:01:00.639Z · LW(p) · GW(p)
I think it will take me another 2-3 months of work to refine my concepts to the point where I will publish them. I would guess that two years would be more likely without access to Gemini 2.5 Pro.
It's typically to not explain valuable new theoretic insights in bits before publishing them properly.
On aspect of how I work is that I have a Google docs in which I write my ideas and regularly give Gemini the new document.
I do define my terms in a BFO-compliant ontology that's in the Google doc. Out of the quest of defining the terms, new questions arise. Gemini can easily check whether there's prior work for the questions.
If you would want to do something similar with AI safety one research direction would be to try to formulate a BFO-compliant ontology for the AI safety discussion on LessWrong. In that quest, you are going to get a lot of open questions. You can ask Gemini whether there's prior work on those questions. I'm right now ordering a philosophy book published in 1986 because it has prior work on a question related to my work.