HAL 9000, the fictional artificial intelligence (AI) character from 2001: A Space Odyssey, or his facsimile, arrived in real life and began to circulate across the web and through social media during the pandemic. In summer 2020, the AI research laboratory OpenAI began offering limited access to a new neural network program called Generative Pre-Trained Transformer 3 (GPT-3), which belongs to a category of deep learning known as a large language model (LLM). LLM programs can answer open-ended complex questions in perfectly composed sentences, generate paragraphs of text and mimic human conversation, à la HAL. A few months after GPT-3 went online, the OpenAI team discovered the AI had taught itself how to program. And in March 2021, OpenAI published a research paper describing the discovery of what it called “multimodal neurons” in GPT-3, inspired by a real class of neurons in the human brain that are activated together in response to general categories or concepts. After more targeted training, GPT-3 now offers Codex, or structured software code in a dozen programming languages. GPT-3 can generate sophisticated legal products, like licensing agreements or leases, which heretofore were the domain of human professionals.
Now, the OpenAI researchers are offering GPT-3 and its LLM platform to developers to explore, tweak and use…for a fee. This is a game changer as it democratizes access to extremely complex and powerful AI. And OpenAI isn’t the only game in town. GPT-3 may be the most well known of the large language models, but Google, DeepMind and Meta all have developed their own LLMs, and Meta became the latest to offer programmers access to theirs for study and use…for free. Having these new resources will catapult research, especially in many medical fields, and further the development of artificial intelligence in increasingly complex practical uses, which is already happening.