Nothing breaks like AI heart

Towards the end of last year you might have heard something about a new AI model called GPT-3. GPT-3 is based on a type of computational model called a neural network. These models are usually trained by throwing a whole bunch of a particular type of data (“screenplays of science fiction movies”, “pictures of apples”) at it and then testing it against versions of itself that gradually get better at predicting an output.

In GPT-3’s case, it has been trained on a very large amount of English language text and is really good at returning original text based on the style, character and tone of whatever you put into it.

I love Pamela Mishkin’s Nothing Breaks Like AI Heart. This interactive essay does an incredible job of showing how GPT-3 works while telling a compelling and personal story (that may be made up by an AI) – give it a read.

Watch this technology jump from your email and document suggestions into legal drafting and policy creation sooner than it should.