3 ms·

Quantify it, show us the numbers.

• NitpickLawyer 2y ago
  I think the field is too new and the successful stories too private atm. However I think the best apples to apples example in this context is Amz's codebase update project that they've blogged about. From memory, they took some old java projects, and had some LLM driven "agents" update the codebase to recent java. I don't know java enough to know how "hard" this task is, but asking around I've heard that "analog" tools for this exist, but aren't that good, bork often, are hardcoded and so on. Amz reported ~70% of code that came out passed code review, presumably the rest had to be tweaked by humans. I don't know if there are any "classical" tools that can do that ootb. So yeah, that's already imrpessive and "available today" so to speak.
  • skydhash 2y ago
    Java is intent as code. It’s so verbose that you have to use an IDE to not go crazy with all the typings. And when using an IDE, you autocomplete more than you type because of all the information that exists in the code
• fragmede 2y ago
  quantifying programmer productivity has been a problem since its inception. lines of code is a terrible metric. so is Jira ticket points. I can tell you that using an LLM, I can make a chrome extension to put a div that says "hello world" at the top of every webpage far quicker than if I had to read the specifications of extension manifests and how to do it manually but how do you quantify that generically? how do you quantify that vs the wasted time because it doesn't understand some nuance of what I'm asking it to do, or when it gets confused about something and goes in circles?
  • skydhash 2y ago
    You could also download a sample extension code, strip out the text and put “Hello World” instead. As fast and no need to train a model to do that.