Let’s set the scene.
You’re a very sociably awkward child, with no friends, but with an unusual imaginary friend born out of your loneliness.
It’s 2002 somewhere in the USA.
This friend, Magnetron, is, in your mind, an English gentleman from the 1900s, who served in World War One, was an immigrant, a poet, and most obviously, an awesome Starcraft player.
With an incredible vivid and elaborate backstory, you would spend hours conversing with your friend. It may have been absurd, but it seemed so real, much to the puzzlement of his parents, but provided much ammunition for teasing by his sisters.
Magnetron was a microwave.
This is a true story.
The child grows up, and has no need for Magentron, and enters the world of tech.
Then, many years later……in fact, a few months ago OpenAI went public with GPT3 (Generative Pre-trained Transformer 3 is an autoregressive language model that uses deep learning to produce human-like text.), and that’s when this man, Lucas Rizzotto, remembered Magentron, and thought, could he bring Magentron alive with AI.
And so, the journey began.
Utilising an Amazon smart microwave, swapping it’s brain with a GPT3 custom solution, equipping it with a microphone and speakers, this custom microwave now could take his voice, send it to OpenAI, and respond in kind.
The only thing left was to provide Magnetron with his “memories”…..you know, a bit like ‘Blade Runner’.
Lucas solved the problem by writing a very elaborate and detailed one hundred page book going into every aspect of Magentron’s imaginary life, from his birth in 1895 all the way through to the time that they first met. He also went into detail regarding a lot of the interactions that they had in the past.
Lucas then informed the GPT3 that all this information was real.
Magentron’s backstory was full of grief, war, and loss; and the GPT3 marked this as important. This became very apparent when Magentron was switched on for the first time.
Magnetron had PTSD!
I’ll not spoil the story, but by all means watch the video of Lucas’ project with Magentron below.
It really does make you think about AI and deep learning; and how important what we input into these systems.