The generative pre-trained transformer 3 (GPT-3) algorithm is a large neural network for natural language applications. With more than 175 billion parameters and trained on over 45 terabytes of data from the internet and books, it has read more than any human writer. This gives it an incredible ability to understand and create natural language outputs based on a brief.
Read the column, and then read their notes from the editor at the end, here…
Just like a human journalist, AIs need an editor – a manager who checks their copy. But with AIs, the supervision goes broader. It’s a key part of the process of working with an AI, regardless of how much training it’s had. Supervision ensures an AI delivers on the practical and nuanced context of the business benefits you need. And it can add back in the ‘common sense’ we talked about here. The notes from The Guardian’s editor show the nature of how AIs could generate multiple articles that could be combined together by a more skilled craftsman. It’s a different approach, but a strong result.
[Editorial disclosure: The Guardian has been a client of our parent company, Digital Strategy Consulting – although the team did not work on this project]