Science is a strong-link problem
4 comments
·February 4, 2025incognito124
Further discussion: https://news.ycombinator.com/item?id=41993801
practal
I like the weak-link/strong-link distinction, but you need to be careful what problem exactly you are talking about. "Science" is not a problem. "How to distribute funding to scientists" is a problem. And here the weakest links actually matter, because you only have a finite amount of funding, and you don't want to waste it on the weak scientists, so you have more for strong scientists.
Another way of phrasing that problem is, "what actually is strong science, and what is weak science"? Over a few decades, that question usually plays out correctly, but even a few years is often not enough to tell, even for specialists, especially for truly new and original science. As an example, right now, would you say abstraction logic [1] is strong or weak science?
And of course, venture capital funding is also not a strong-link problem. You can waste money on some weak startups that you thought were actually strong, but do that too much, and your fund goes under.
brudgers
I read this and thought, I wonder where LLM's fall in the range...I think about the quality of LLM's more than I want.
One purpose of science is to provide the rest of the society, who are not scientists, with reliable insights, ideally with actionable advice on how to solve a problem. If you treat science purely as a stong-link problem, the burden of quality control lays with the consumer of science. Peer review attempts to lay it with experts. That approach is nowhere near perfect, but is the best we have. And it scales much better.