3 ms·

GitHub CEO says Copilot will write 80% of code “sooner than later”

infamouscow 3y ago
If 100% of code Copilot generates is potentially laundered AGPL code, the 80% of the code Copilot writes is still be useless for most businesses.
stefanos82 3y ago
If that's the case, then what's the point dealing with technology anymore? I do so because I love fixing problems. If I can't feel I have some kind of value, why should I compete with a machine that gets smarter and smarter with every iteration it goes through various models?
SavageBeast 3y ago
Humans copy/paste/modify "boiler plate" code all the time. Absolutely no reason an AI based system couldn't do the same. Is 80% of all code "boiler plate"? Maybe, depending on how you define it. There are only so many ways to collect data from a form for example. Optimistic predictions of grandeur from CEOs of companies isn't a new thing either. AI will certainly automate the current manual process of identifying the code you need on some internet site and duplicating it for your use case by copy/paste/integrate. Maybe the up-shoot of this will be a new higher level library of Common Things Every Codebase Must Do. That seems to be what we're really talking about here. If 80% of code is mostly boiler plate and only 20% of it is Real Complexity this really isn't too remarkable. The Real Problem here will be the same as its ever been; in order for AI to write proper solutions it will need good prompts telling it what to do right? Thats requirements. Solid, accurate, concise requirements. Such things don't exist in my 25 or so years as a Corp. America Software Developer. In fact, great lengths have been gone to such that "requirements" are no longer strictly necessary. Consider the old transition from Waterfall to Agile in this context. The whole thing of Agile in my opinion is the pushing down of requirements on dev teams to figure out as they go. Garbage requirements make garbage implementations. I don't see how AI is going to fix this any time soon.

If 100% of code Copilot generates is potentially laundered AGPL code, the 80% of the code Copilot writes is still be useless for most businesses.

stefanos82 3y ago

If that's the case, then what's the point dealing with technology anymore? I do so because I love fixing problems. If I can't feel I have some kind of value, why should I compete with a machine that gets smarter and smarter with every iteration it goes through various models?

SavageBeast 3y ago

Humans copy/paste/modify "boiler plate" code all the time. Absolutely no reason an AI based system couldn't do the same. Is 80% of all code "boiler plate"? Maybe, depending on how you define it. There are only so many ways to collect data from a form for example. Optimistic predictions of grandeur from CEOs of companies isn't a new thing either. AI will certainly automate the current manual process of identifying the code you need on some internet site and duplicating it for your use case by copy/paste/integrate. Maybe the up-shoot of this will be a new higher level library of Common Things Every Codebase Must Do. That seems to be what we're really talking about here. If 80% of code is mostly boiler plate and only 20% of it is Real Complexity this really isn't too remarkable. The Real Problem here will be the same as its ever been; in order for AI to write proper solutions it will need good prompts telling it what to do right? Thats requirements. Solid, accurate, concise requirements. Such things don't exist in my 25 or so years as a Corp. America Software Developer. In fact, great lengths have been gone to such that "requirements" are no longer strictly necessary. Consider the old transition from Waterfall to Agile in this context. The whole thing of Agile in my opinion is the pushing down of requirements on dev teams to figure out as they go. Garbage requirements make garbage implementations. I don't see how AI is going to fix this any time soon.