Many years ago (early 2000s), I had to write a tool to scrape Yahoo message boards. The business was that folks were running "pump and dump" scams on the finance boards. The companies whose stock was being "pumped" hired law firms who, in turn, hired the company I worked for.
I was VERY new to Perl and didn't realize that LWP::simple already existed. I therefore ended up writing my own library using TCP socket handling and sending GET requests "by hand".
It was a great learning experience and taught me a lot about how message boards, TCP and HTTP work. At the same time, it was slow, took a lot of time and had limited features and very little error handling.
I now use Python's requests module all the time and have never, not ever, thought "I should go peak inside the library to see how it actually works under the hood".
My point in this story is that LLMs will probably move us more and more towards "AI as library". Sure, if you are writing super higher performant code that ties tightly to hardware you might still dig down into the details.
Most of us will probably just use the next generation "library".
Tricky though because this generation of LLMs involve random sampling, so unlike computer code and libraries it's non-deterministic and inherently unreliable. Not sure if autoregressive random token sampling will ever be the winning paradigm.
I was VERY new to Perl and didn't realize that LWP::simple already existed. I therefore ended up writing my own library using TCP socket handling and sending GET requests "by hand".
It was a great learning experience and taught me a lot about how message boards, TCP and HTTP work. At the same time, it was slow, took a lot of time and had limited features and very little error handling.
I now use Python's requests module all the time and have never, not ever, thought "I should go peak inside the library to see how it actually works under the hood".
My point in this story is that LLMs will probably move us more and more towards "AI as library". Sure, if you are writing super higher performant code that ties tightly to hardware you might still dig down into the details.
Most of us will probably just use the next generation "library".