No matter how much I code and how much I continue to cross barriers and prove that all of this is very much doable with the right intentions and tools, I don’t think I’ll ever be a ”so I’ll write...
No matter how much I code and how much I continue to cross barriers and prove that all of this is very much doable with the right intentions and tools, I don’t think I’ll ever be a ”so I’ll write up a file system to check this rare bug “ programmer.
Writing a FUSE FS is honestly surprisingly easy, especially if you need a simple thing for testing purposes and not a production ready FS with great error handling and stuff. You basically just...
Writing a FUSE FS is honestly surprisingly easy, especially if you need a simple thing for testing purposes and not a production ready FS with great error handling and stuff. You basically just have to write a bunch of simple boilerplate code (or ask AI to generate it, as the author did) to implement filesystem methods.
I wrote a proof of concept joke overlay filesystem which stores an LLM-generated summary of the file instead of the actual content and then tries to generate the file back when it's read, so you'll get different content every time you read the same file. It was less than 250 lines of Python: https://github.com/flexagoon/morok_fs/blob/main/morok_fs.py
And it's useful in some contexts too! Like for versioning things to ensure that you know who leaks/breaks an embargo or similar. You don't have to write all those versions now, just save them and...
And it's useful in some contexts too! Like for versioning things to ensure that you know who leaks/breaks an embargo or similar.
You don't have to write all those versions now, just save them and distribute.
I was pleasantly surprised that the author was so candid not only on their use of generative AI but that they wouldn't have even made the FS without Claude. I really think that's the future of the...
I was pleasantly surprised that the author was so candid not only on their use of generative AI but that they wouldn't have even made the FS without Claude.
I really think that's the future of the profession. Being broadly knowledgeable about what tools exist without having to know the ins and outs.
An example I had recently was having gen AI make a bash script to help debug a really annoying dependency issue. It was the sort of thing where, if I had to write the code, I probably would have just done things manually. But since I knew the tools available, all I had to do was ask to have them employed for me, if that makes any sense.
I'll also briefly point out this is Jane Street, just about the highest paying programming job you can get. This author is indeed in a different stratosphere from literally 99.99% of even professional developers.
The language model when intelligently prompted, yes. There is in fact a human that does need to know about the broad topics at play, but it has never been efficient to have to know something...
The language model when intelligently prompted, yes. There is in fact a human that does need to know about the broad topics at play, but it has never been efficient to have to know something inside and out.
We only have limited time on this earth, I will never learn almost the entire field. I don't see how it's a bad thing at all that I'm enabled to work and play with concepts and technologies I would never have dreamed of touching before my time is up.
Very fascinating! I'm so grateful when companies take the time to do these sorts of detailed postmortems. Coming at it from the perspective of a real, production issue makes them so enlightening.
Very fascinating! I'm so grateful when companies take the time to do these sorts of detailed postmortems. Coming at it from the perspective of a real, production issue makes them so enlightening.
No matter how much I code and how much I continue to cross barriers and prove that all of this is very much doable with the right intentions and tools, I don’t think I’ll ever be a ”so I’ll write up a file system to check this rare bug “ programmer.
Fantastic work.
Writing a FUSE FS is honestly surprisingly easy, especially if you need a simple thing for testing purposes and not a production ready FS with great error handling and stuff. You basically just have to write a bunch of simple boilerplate code (or ask AI to generate it, as the author did) to implement filesystem methods.
I wrote a proof of concept joke overlay filesystem which stores an LLM-generated summary of the file instead of the actual content and then tries to generate the file back when it's read, so you'll get different content every time you read the same file. It was less than 250 lines of Python: https://github.com/flexagoon/morok_fs/blob/main/morok_fs.py
That idea is fucking hilarious! Like "I Am Sitting In A Room" but for every file on your computer hahah
And it's useful in some contexts too! Like for versioning things to ensure that you know who leaks/breaks an embargo or similar.
You don't have to write all those versions now, just save them and distribute.
I was pleasantly surprised that the author was so candid not only on their use of generative AI but that they wouldn't have even made the FS without Claude.
I really think that's the future of the profession. Being broadly knowledgeable about what tools exist without having to know the ins and outs.
An example I had recently was having gen AI make a bash script to help debug a really annoying dependency issue. It was the sort of thing where, if I had to write the code, I probably would have just done things manually. But since I knew the tools available, all I had to do was ask to have them employed for me, if that makes any sense.
I'll also briefly point out this is Jane Street, just about the highest paying programming job you can get. This author is indeed in a different stratosphere from literally 99.99% of even professional developers.
But then who actually knows the ins and outs? Just the language model? The fancy translation machine?
The language model when intelligently prompted, yes. There is in fact a human that does need to know about the broad topics at play, but it has never been efficient to have to know something inside and out.
We only have limited time on this earth, I will never learn almost the entire field. I don't see how it's a bad thing at all that I'm enabled to work and play with concepts and technologies I would never have dreamed of touching before my time is up.
Very fascinating! I'm so grateful when companies take the time to do these sorts of detailed postmortems. Coming at it from the perspective of a real, production issue makes them so enlightening.