Talking to computers

Key Takeaways

We used to have to learn computer languages, but thanks to AI, computers are now learning to speak our languages. In 2017, AI researchers developed a new architecture to train computers to process text in natural languages like English, Japanese or Arabic. The architecture called a transformer enables machine learning models to develop a more robust understanding of language. Older natural language techniques were unable to clarify word meanings which humans could infer from context and basic reasoning. However, during model training, transformers used a technique called self attention to identify relevant relationships between tokens such as words in a passage of text. By applying this technique to vast amounts of text, these large language models develop sophisticated natural language capabilities that we can apply to a multitude of tasks from summarizing business reports to explaining mathematical calculations step by step and even generating the right Python code to solve a problem. To learn more about how large language models work, as well as a variety of other machine learning topics, visit machine learning crash course.