AI to Generate Text and Music
Here are five important AI stories from the week.
Earlier this year, OpenAI introduced an improved language model, known as GPT-2, one potentially so powerful at generating new text that OpenAI chose NOT to release the model.
However, OpenAI did release a much smaller version of GPT-2. The Allen Institute released a working demo of this smaller model; play around with the demo and just see for yourself how impressive these new language models are at generating new text.
We are in the early stages of big breakthroughs in the applied NLP space.
More musicians are incorporating AI into their music-making process, viewing AI as an enabler of creativity rather than as a direct threat. Here’s a great read on just how artists are using machine learning applications to create new music. This is a preview of the type of human-AI collaboration we will see across the creative fields in the coming years.
This article—and the book it references, Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass—sheds light on the humans that are behind the “AI” services available today such as Amazon Alexa and Google Duplex. Most AI today is not 100% machine-driven; rather machines do the easy work, the humans do the hard work, and the service is packaged as AI.
Machine learning is notoriously black box-y, making it difficult for both technologists and business people to understand which models are performing well and why. In the past year, considerable progress has been made to shed light on what happens under the hood of machine learning models. MIT discusses visualization as one possible solution. The startup Weights and Biases offers similar visualization solutions for deep learning.
China and the U.S. are locked in a very competitive AI race, and this read does a great job digging into what types of AI and which players are dominating in China. Chinese news on AI is often inaccessible to English speakers, but the ChinAI newsletter does a good job providing translations for English-speakers.
More Stories Worth Reading and Watching…