Code is plentiful
And there is a gigantic quantity of code on the market. Simply consider GitHub alone. A back-of-the-envelope calculation says there’s round 100 billion traces of open-source code accessible for coaching AI. That’s a whole lot of code. A complete lot of code.
And in the event you want an evidence of how code works, there are one thing like 20 million questions and much more solutions on Stack Overflow for AI to study from. There’s a purpose that Stack Overflow is a shell of its former self—all of us are asking AI for solutions as an alternative of our fellow builders.
Code is verifiable
As well as, code is well verified. First, does it compile? That’s at all times the massive first check, after which we will test by way of testing if it really does what we wish. Not like different domains, AI’s code output may be checked and verified pretty simply.
In case you select to, you may even have your AI write unit and integration checks beforehand, additional clarifying and defining what the AI ought to do. Then, inform your AI to write down code that passes the checks. Ultimately, AI will work out that test-driven improvement is one of the best path to writing good code and executing in your needs, and also you received’t even should ask it to do this.
Welcome, Skynet
And at last, code is a good use case for AI brokers as a result of builders are typically unafraid of recent know-how and at all times appear able to check out a brand new instrument. This turns into a virtuous circle as AI firms produce coding brokers, and builders embrace these coding brokers. Software program improvement is a large a part of the financial system, and AI firms are strongly incentivized to lean into profitable markets which might be accepting and keen about utilizing AI brokers.
