Hacker News Reader

> the DeepMind team built a custom dataset from CodeContests from two previous datasets, with over 13,500 challenges. Each came with an explanation of the task at hand, and multiple potential solutions across multiple languages. The result is a massive library of training data tailored to the challenge at hand.

Isn't this just over-fitting the model?

View on HN

Isn't that exactly how the vast majority of engineers pass interviews today? They study up on the handful of algorithms that come up in interviews and then (with varying levels of acting) recite the answer to the question they knew they would get. What? Overfittings only bad in if you're a robot huh?

It tickles me that organic neural nets that learn by viewing information are incredulous that neural nets could learn by viewing information.

Yeah the article mentions that they outperformed Codex, but was Codex specifically trained on code contests?

Reply on news.ycombinator.com