Scientists and technologists have been using AI for decades. We’ve used it to do complicated calculations and run algorithms and equations that we couldn’t previously conceive of. Your favorite streaming services have been using it for years to recommend shows and movies. But looking at media coverage of the past year, you’d think that AI was just developed. Why is mainstream AI language processing now taking off?
In late 2022, did AI experience an onslaught of media attention that made it seem like it was a new functionality? Why are legislators and regulators now racing to regulate something that has been in existence for about the same length of time as the color TV?
Learning To Learn
Tools powered by AI have essentially learned to learn. The language models we’re all seeing now train themselves with two primary algorithms. First, they can look at any sentence in any context and try to predict the next one.
The other way that language models try to learn is by guessing words in a sentence if some words are randomly removed. These are examples of implicit supervised training, and it’s made possible because these tools use the entire corpus of the internet as training data. This is the actual breakthrough.The other way that language models try to learn is by guessing words in a sentence if some words are randomly removed. These are examples of implicit supervised training, and it’s made possible because these tools use the entire corpus of the internet as training data. This is the actual breakthrough.
