According to https://spectrum.ieee.org/chatgpt-for-coding, ChatGPT is anywhere from 89% to .66% accurate when writing computer code. The large swing is based on the difficulty of the task and the requested language. The largest factor however, was whether or not the requested task was in the training data for that particular model. If the task is something completely novel, ChatGPT’s success rates dropped to the 0.66% level.
At the same time, they found that the ChatGPT written code was typically 50% smaller and faster than human written code.
This analysis makes it clear that its going to be a while before human coders are completely replaced by Generative AI models, but that there is a place for AI created code as long as there is a human around to help it across the finish line.
#AI #ai coding #chatgpt #coding #LLM