OpenAI, the company behind ChatGPT, recently released a podcast on its YouTube channel. The first episode features CEO Sam Altman.
While many would tune in to hear Altman share his vision for the future of his company and artificial intelligence, the most intriguing part of the episode might be the identity of the person hosting him.
The host, Andrew Mayne, spent over three years from 2020 to 2023 as a technical staff member at OpenAI, where he played a key role in the development of ChatGPT.
A quick Google search would also reveal that he took one of the most unconventional paths to becoming an essential part of today’s AI revolution. Before joining OpenAI, he was a magician, illusionist, and best-selling author.
None of these were side hustles for Mayne—he had his own show on A&E Networks, a special on the Discovery Channel, and was a best-selling author of thrillers and mysteries, according to The Wall Street Journal.
Cut to the present day, after his time at OpenAI, he now runs an AI consulting firm. So, how did he end up on this unexpected path?
A Star on Screens, Books, and Classrooms
Mayne’s professional career as a magician began in his teenage years, when he performed in resorts and on cruise ships.
Soon, he started his first world illusion tour. Moreover, at the peak of his career in the 2010s, he starred in the show ‘Don’t Trust Andrew Mayne’ at A&E Studios, having invented over 400 magic and illusion effects.
However, Mayne’s talents extended far beyond live performance. His distinguished career in magic also included writing 45 books on the art, and he collaborated with big names in the industry, such as David Blaine and David Copperfield.
In 2011, he made another creative leap, venturing into fiction writing with his debut novel ‘Angel Killer’.
This transition proved remarkably successful—he was deemed a Wall Street Journal best-selling novelist and an Edgar and Thriller Award finalist. One of his books also stood for six weeks at the number one spot for all books on Amazon.
Yet, there was another dimension to Mayne’s career that would prove crucial to his work on AI later. Besides being a magician, illusionist, and author, Mayne was also a science communicator. With support from the Johnny Carson Foundation and the James Randi Educational Foundation, Mayne created a programme to use magic to teach critical thinking skills in public schools.
Moreover, his ‘Wizard School’ segments, which taught magic and science to children, aired nationwide on public television.
But most importantly, technology has swept into every aspect of his diverse career path.
Magic With Machine Learning
In an interview with The Creative Penn, Mayne stated that he was always interested in the intersection of science and entertainment as a kid.
The seeds were sown early when he got his first computer and started building chatbots. He built simple chatbots that used variables to remember input details and reproduce them when needed.
Some of Mayne’s magic tricks even involved technology, and he performed numerous tricks with the original iPhone. He even built a prank site to trick people into believing their iPhone was broken, and this was before the App Store was even launched.
What came next would be the bridge between his entertainment career and his future at OpenAI. When Mayne decided he was done with TV and wanted to enter the world of technology, he set out on an ambitious project at the intersection of magic, illusion, science, and technology.
In a special episode for the Discovery Channel, he constructed a suit to make himself invisible to sharks—essentially, to swim close to a great white shark without being noticed, or even better, uneaten.
Mayne detailed his experience in an interview with Logan Kilpatrick, a group product manager at Google DeepMind.
He engaged with multiple scientists to understand the senses of a great white shark and how they perceive their surroundings.
Mayne then built a shark camouflage suit using LEDs to match the silhouettes of the front and back and a 360° shark detector using OpenCV that would alert him. He decided to use a 3D engine image, a video of a great white shark, and gathered thousands of screenshots from all possible angles as data for the algorithm.
His plan was a success.
“As I learned more about shark vision and the correlations with machine learning, I realised this was something I want to pursue full time,” Mayne said in a LinkedIn post.
The Path to OpenAI
Building this shark camouflage suit led him down the path of AI, but he also wondered how this technology would come into the picture in his career as an author, where he was writing two mystery thriller novels a year. As someone whose livelihood depended on creative writing, he tells Kilpatrick that he was curious to find out how long it’d take before AI came for his job as a writer.
And back then, OpenAI was working on the very first version of their GPT systems.
He began experimenting with next-word prediction AI models. Notably, Mayne stated that he was simply curious about these models and was unaware that GPT-1 had even been released.
“But when GPT-2 came out, that was a big deal because it was the first time you saw a text model that could produce coherent content for more than a few sentences,” he said.
Mayne then read every document in the repository of content written by GPT-2, which OpenAI had published on GitHub. This led him down the rabbit hole of transformers, attention mechanisms, and more.
His deep dive into the world of AI language models proved worthwhile when he received a call from OpenAI.
Ashley Pilipiszyn, the company’s then-technical director, contacted Mayne after hearing him speak on a podcast about AI and GPT. Mayne was invited to help the company build GPT-3.
He said yes to the idea, and in Mayne’s own words, he went “nuts nonstop day and night” experimenting with GPT-3, using various prompts, giving feedback to OpenAI, and helping them discover novel capabilities of their models. He was running a project called ‘Creative Applications’, and was working with different teams and researchers.
Subsequently, he wrote a substantial portion of the ‘prompt library’ included in OpenAI’s documentation. Mayne revealed that he created many of the original prompts and examples that are used today.
Moreover, Mayne personally developed AI chatbots, utilising techniques to search through data repositories to answer user questions and support long conversations, such as what’s now called retrieval augmented generation (RAG), well before these methods became popular as they are today.
He also built a social network platform called ‘AI Channels’ that lets users talk to AI chatbots with specific personalities and styles.
He then went on to work as a science communicator for OpenAI, besides being the company’s first prompt engineer.
Moreover, he has extensively written about AI, targeting both the average user and developers on how to make the best use of these models. He also has a GitHub repository with numerous projects.
After a successful stint at OpenAI, Mayne now runs Interdimensional, an AI consultancy firm that helps companies build and deploy AI solutions.
The post When OpenAI Hired an Illusionist and a Magician to Build ChatGPT appeared first on Analytics India Magazine.
