“Paper without code” is an AI project to provide prototype implementations for classic and trending AI papers by using LLM code generation.
Q: “Papers With Code” exists, why do we still need “PaperWithoutCode”?
Short answer: There are numerous groundbreaking papers without accompanying code, and as researchers, we should understand and digest these valuable contributions as well.
Long answer: As a senior AI researcher who frequently engages with the latest developments in our field, I’ve often pondered this question. Let me share my thoughts from both a reader’s and an author’s perspective.
As a reader, I find myself sifting through countless arXiv papers each week, searching for innovative and inspirational AI research. Naturally, I tend to prioritize papers with code implementations, as they offer immediate practical insights. However, this approach has led me to question whether I might be overlooking exceptional work simply because the code isn’t readily available on publication day.
On the flip side, as an author, I understand the challenges of releasing code alongside a paper. Producing a complete, bug-free implementation that’s well-organized, optimized for efficiency, and thoroughly tested for edge cases requires significant additional effort. Moreover, navigating the complexities of legal approval for code release within a corporate environment can be time-consuming and sometimes prohibitive.
The crux of the matter is this: the primary value of a research paper lies in its unique contribution to our collective knowledge. Our evaluation and selection of papers shouldn’t be biased by the presence or absence of accompanying code. The ideas, methodologies, and theoretical advancements presented in a paper can be groundbreaking and influential, regardless of whether code is immediately available.
This realization led me to initiate the “PaperWithoutCode” project. By leveraging LLMs to read papers and generate code for educational purposes, we can bridge the gap between theoretical concepts and practical implementation. This approach allows us to explore and learn from a wider range of research, ensuring that valuable contributions aren’t overlooked simply due to the absence of immediate code availability.
In essence, while “Papers With Code” serves a crucial role in our community, “PaperWithoutCode” complements it by ensuring we don’t miss out on potentially transformative ideas that may not yet have a public implementation. As researchers, it’s our responsibility to engage with and learn from the full spectrum of work in our field, code or no code. Please suggest good papers by submitting at https://github.com/phunterlau/paper_without_code