Member-only story
AlphaCode From DeepMind Surpassed Humans at Coding
Heard of AlphaCode? It’s a new AI model developed by DeepMind, and it’s taking the programming world by storm.
In a recent competition called CodeContest, AlphaCode outperformed around half of human programmers, solving 30% of some of the world’s hardest coding challenges, given them a right run for their money. And that’s just the beginning.
In this contest, coders were presented with with challenges written in natural language form, requiring them to write a program to complete the task as quickly and accurately as possible.
A step-change in AI performance
30% may not seem like a high success rate, but remember that these were highly complex problems that even the most advanced human programmers on the face of the Earth would find difficult to crack.
To put this into perspective, when a similar challenge was presented to OpenAI’s Codex model, albeit impressive in its own right, it only achieved single-digit success.
A data-only learning method
Unlike other AI programming models, AlphaCode doesn’t have any built-in knowledge or pre-conceived notions about computer code. It’s starting from a blank slate, not knowing anything about code syntax or structure, and it…
