A thoughtful post on the limitations on deep learning: https://arxiv.org/pdf/1801.00631.pdf
I think the author's criticisms of deep learning's coverage in the media are fair, and that we should consider it another tool in the toolbox. Comparing it with AGI seems like a very high bar though.
miniblog.
Related Posts
Python's GIL is well known, but many languages have limitations on multithreading (e.g. JS, Ocaml).
Perhaps this is a consequence of C as a popular implementation language? A GIL could be less common when Rust is the implementation language.
The effectiveness of fuzzing, the limitations of current research, your best options today, and a worked example:
physical-cpu-count is a lovely example of small packages on npm doing one thing well. It's a cross-platform solution with clearly discussed limitations and design decisions regarding hyperthreading.