3 ms·

I remember reading that reproducibility is hard in these scholarly articles... When will an AI attempt to reproduce the results from all CS and math articles.

• visarga 4y ago
  That's an interesting expectation, the models we have today reach average human level but few can surpass the top human in a task (like AlphaGo). But who would pay for the massive compute necessary for such a large scale task?
  • hackernewds 4y ago
    Presumably an AI will reflect fewer than the creators intellect., made worse by the collaboration losses necessary to accomplish the project now
• closedloop129 4y ago
  I am not quite sure if that's your point but math articles, and CS articles to some extend, don't rely on experiments to the same extend as other fields. A mathematical proof doesn't have to be reproduced but verified. It will be a huge step forward when an AI can automatically verify proofs beyond the already used proof systems but that should be unrelated to reproducing experiments. What I am missing are open source repositories, docker images and virtual machines for all CS papers. Why is it acceptable to publish papers without enabling everybody else to instantly reproduce it when it is technically possible?