-
Book Overview & Buying
-
Table Of Contents
Supercharged Coding with GenAI
By :
Statistical and AI approaches can learn patterns from sequential data such as text or code, which can efficiently predict the next symbol. Due to the range of possible words, deep learning methods scale well to efficiently predict the next words much better than statistical techniques and have become the standard. Even so, they require vast training sets of text and code and an expensive computational process to be effective.
While LLMs are highly effective, they are not perfect and are limited by their training set of data and perhaps the time it was trained. The output can be limited for various domains and time periods, which can result in hallucinations. While research continues in these areas, some solutions to improve results include effective prompt engineering, providing contextual information such as documents or examples, fine-tuning, and the RAG approach.
OpenAI and ChatGPT are trained on general data and code, and are quite effective at generating code...