Fallacies of AI Driven Coding
A few days ago, DeepMind (acquired by Google in 2014) releasedAlphaCode and self-publishedapaperexplaining how their artificial intelligence (AI) can ���understand���a programming contest task written in English and then write a Python, Java or C++ program, whichwould work in about 30% of cases.Earlier last year OpenAI ($1B-funded by Microsoft in 2019)released Codexand published a paper, claimingthat their AI can also solve around 30% of the programming tasks it wastested with.Wire...
date
newest »
newest »
I love that a coder like yourself is pushing back on these claims. As is typical of the AI futurism, the claim that learning models can "write code" is based on a fuzzy logic, very similar to the claim that NLP models "read." It will in fact regurgitate prior patterns, and so is a very poor candidate for intelligent tasks like writing code. There may of course be breakthroughs here in the future, but it won't be from the current big data approach.
Best,
Erik