OpenAI-announced Language Models are Unsupervised Multitask Learners

An impressive new language AI writes product reviews and news articles. Its creators are worried about misuse.

The OpenAI team demonstrated that we could get those results from an “unsupervised” AI — meaning the system learned from reading 8 million Internet articles, not from being explicitly trained for the tasks.

The AIWS model includes implementations to reduce and monitor harmful things, misuse, and slanders that AI can unleash onto the world.