Sébastien Bubeck, a machine learning researcher at Microsoft, was inspired by Artificial Intelligence (AI) when he woke up one night last September. He was recently given access to GPT-4, an advanced language-generation algorithm from OpenAI, and was part of a team working to integrate this technology into Microsoft’s Bing search engine. Bubeck and his colleagues were impressed with how extraordinary GPT-4 seemed when compared to their prior experience with AI.
GPT-4 was trained using vast amounts of data and code, enabling it to make statistically plausible guesses when presented with certain text. One night, Bubeck tested the system by asking it to generate a unicorn image using TikZ, a programming language for producing scientific diagrams. Unbelievably, the code that GPT-4 provided actually produced a crude yet recognizably unicorn-like form from a series of ovals, rectangles, and a triangle. Bubeck was convinced that, in order to succeed with this task, GPT-4 must have a certain level of intelligence.
With OpenAI releasing their chatbot ChatGPT, powered by GPT-3, last November, the world was stunned by its numerous talents, such as writing poetry, fixing coding issues, and accumulating knowledge from a range of sources. While people may feel a sense of amazement, there is also legitimate concern regarding the potential for economic disruption, cheating, and misinformation.
Researchers are now trying to gain a better understanding of AI’s capabilities, and are only just starting to analyze the technology behind ChatGPT and similar systems. Nonetheless, Microsoft released a paper with very early findings that indicated GPT-4 had the capacity to do more than just make statistically logical decisions — proving that it had sparks of Artificial General Intelligence. The paper featured examples to showcase how GPT-4 could perform varied tasks, rather than focusing on a single area like prior AI programs such as GPT-3. This hints to the potential of GPT-4 developed further in the near future.
Microsoft is a leading multinational technology company headquartered in Redmond, Washington. It is best known for developing, manufacturing, licensing, and supporting a wide range of technologies including computer software, personal computers, consumer electronics and related services. Since its founding in 1975, Microsoft has gained a reputation for technological innovation and has become the world’s largest software company with a $1.38 trillion market capitalization in 2019.
Sébastien Bubeck is a machine learning researcher at Microsoft. Before joining Microsoft, he was an AI researcher at Princeton University, working on topics such as online learning and optimization. In 2009 he was awarded the LEO Prize in Online Learning, and has published numerous papers on machine learning algorithms and artificial general intelligence. He currently focuses on developing “smarter” AI at Microsoft, particularly through integration of GPT-4 into Bing. He is seen as a pioneer in the development of AI technology and AI-powered search applications.
