Yep, I pretty much stopped using Twitter entirely.

If you don't want to see garbage, stop looking at my GitHub

Big models are cool, but I really like how the mini models get better and cheaper. Opens up a lot of options

The Internet is just a fad

I suspect advances in AI debugging capabilities will come from new ways of capturing, organizing, and retrieving relevant information. And I don't think it will be sufficient to just gather lots of unstructured data, throw it at a vector DB, and call that "knowledge"

I remember a debugging guru explaining to me that one particular crash was debugged not just by looking at the assembly code, but also by looking at the region where the crashes originated. And realizing that region pirated Windows a lot. And pirated windows came with certain malware... that crashed

But what about those 1% of cases that require deep analysis? Will an LLM help with those cases? I hope so. But I also suspect the biggest challenge there is the collection, organization, and retrieval of domain knowledge that can't be directly learned from the source code or a stack trace.