OpenAI

OpenAI is an American technology company that specializes in artificial intelligence (AI). The company is known for its Chat Generative Pre-trained Transformer, or ChatGPT, a language model chatbot that creates computer-generated content that mimics human conversation. Following the release of ChatGPT in November 2022, the company became one of the most valuable technology start-ups worldwide, with its worth estimated to be at least $80 billion by early 2024.

Background

One area in which artificial intelligence (AI) research has made, and continues to make, technological strides is in the development of advanced generative language models. The goal of language models is to intake human communications, commonly in the form of text, and output an understandable, logical, truthful, and/or correct response that maintains the context of the situation. In the 2020s, after several versions of language models, large language models (LLMs) were dominating the field, and OpenAI had become a leader in developing LLMs and AI technology.

LLMs use web crawlers and a variety of sources and databases to glean hundreds of billions of bits of textual and coding communication from the web. These data are then used to train the model to correctly discern an appropriate response. OpenAI’s GPT-n series of LLMs is a market leader in LLMs and powers its popular chatbot ChatGPT.

Overview

OpenAI is on the cutting edge of artificial intelligence (AI) research. The company was founded in December 2015, and is headquartered in San Francisco, California. OpenAI’s stated mission is to develop artificial general intelligence (AGI) for the good of all humanity. AGI is a yet-unrealized system that can outsmart the human brain—it can, in some sense, think for itself. While artificial narrow intelligence, sometimes known as weak AI, focuses on a specific task or tasks (for example, playing chess or Go, operating machinery, writing code, etc.), AGI can perform general tasks without being trained specifically to do so. Some have suggested that AGI’s GPT-4 pushes the boundaries of AI, perhaps edging into AGI territory.

OpenAI was cofounded by Ilya Sutskever, Greg Brockman, Sam Altman, tech billionaire Elon Musk, and several other notable figures in Silicon Valley. Brockman served as the company's CEO until 2019, while Sutsekever held the position of chief scientist. Altman replaced Sutskever as CEO in 2019. Musk initially promised a huge investment in the company and was a member of its board of directors. However, after coming into conflict with Altman and other executives on the direction of the company, he left in 2018. Musk was confident that Google was far outpacing OpenAI’s work, especially after it purchased the AI research firm DeepMind in 2014. Musk continued to speak out against the company in the following years, arguing that it should be open-source, among other disagreements. In July 2023, Musk announced the formation of his own AI research firm, xAI.

At first, OpenAI asked for “donations,” knowing that any money invested would be sucked up by the high price of AI development and model training. However, after the departure of Musk, along with his money, the company decided to create a for-profit subsidiary called OpenAI Global. In 2019, Microsoft backed OpenAI with a $1 billion investment and returned in 2023 with another $10 billion infusion of funds.

OpenAI’s primary focus has been on the development of its GPT-n series of language models. GPT stands for Generative Pre-trained Transformers. GPT’s systems are trained with so much data from webpages, social media, Wikipedia, books, etc., that they can create text that appears as if it was written by humans. These natural language processing tasks can answer questions, translate language, summarize text, and so on.

OpenAI released GPT-1 in 2018 with 117 million parameters, far outpacing all previous language models. GPT-1 was trained using CommonCrawl, a web-crawling database with billions of words, and BookCorpus, a database that includes more than 11,000 books. GPT-2, launched in 2019, improved the ability to process natural language. Using data from CommonCrawl and WebText, its parameters totaled 1.5 billion. It was a significant improvement over GPT-1, but it still struggled to maintain meaning and context beyond a short paragraph or snippet of text.

In 2020, OpenAI exponentially increased the power of its LLM with the release of GPT-3. This iteration included 115 billion parameters—about 100 times that of GPT-1 and nearly ten times that of GPT-2. It was trained with data from a variety of sources, including CommonCrawl, BookCorpus, and Wikipedia, totaling roughly one trillion words. GPT-3 was much better at understanding and responding to context and could even create poetry. It powered OpenAI’s chatbot, ChatGPT, which in August 2023 had more than 100 million users. Although considered cutting-edge technology, GPT-3 had flaws. It sometimes failed to understand context and provided faulty, misleading, or nonsensical responses. OpenAI released an improved 3.5 version prior to its release of GPT-4 on March 14, 2023.

GPT-4 was another significant improvement of the previous version, boasting 1.7 trillion parameters. Its other advantages included the ability to accept longer passages of text (up to 25,000 words) from users, process images, and perform creative tasks. For example, an OpenAI engineer fed GPT-4 a hand-drawn model of a website, and GPT-4 was able to accurately write the code to create it. GPT-4 could also write original poetry, song lyrics, and fables. While exciting, these abilities led to concerns, such as the rapid pace of AI development, intellectual property issues regarding the information ChatGPT drew on for its machine learning, and the use of ChatGPT for plagiarism or other forms of academic dishonesty. For example, a group of high-profile authors, including George R. R. Martin, Jonathan Franzen, Jodi Picoult, and John Grisham, filed a class-action lawsuit against OpenAI in late 2023. The authors alleged that the company had used their copyrighted works without their consent in training ChatGPT. The New York Times and other news outlets also filed lawsuits against OpenAI, citing similar concerns.

In November 2023, OpenAI's board of directors pushed Altman out of the company, citing a lack of transparency in communications. Brockman immediately announced his resignation as well. Employees of OpenAI were shocked by the decision and quickly signed a letter calling for Altman and Brockman to be reinstated. Although it was reported that the two would be moving to Microsoft to lead an AI research team there, four days after the news first became public, Altman and Brockman returned to OpenAI and a new board of directors was named. In early 2024, Altman helped orchestrate a deal that resulted in tripling OpenAI's worth to approximately $80 billion.

In May 2024, OpenAI announced an updated model called ChatGPT-4o, which promised faster results and improved text and audio capabilities, including a new conversational AI system called "Sky." However, the company encountered controversy following its release of the update. Actor Scarlett Johansson accused OpenAI of using a likeness of her voice in its new voice assistant, despite reportedly telling Altman multiple times that she was not interested. In response, OpenAI removed the voice and hired a legal team to manage its growing legal challenges. Later that month, Sutskever announced his departure from the company. He noted that he regretted his decision to oust Altman from the company the previous year.

OpenAI also released DALL-E, a system that can turn natural language text into realistic images with specific styles, concepts, and attributes. It became publicly available in November 2022. It announced a new tool called Sora in February 2024 that could create videos from users' prompts.

Bibliography

Ali, Fawad. “GPT-1 to GPT-4: Each of OpenAI’s GPT Models Explained and Compared.” Make Use Of, 11 Apr. 2023, www.makeuseof.com/gpt-models-explained-and-compared/. Accessed 8 Aug. 2023.

“Creating Safe AGI That Benefits All of Humanity.” OpenAI, openai.com/. Accessed 24 May 2024..

Hao, Karen. “The Messy, Secretive Reality Behind OpenAI’s Bid to Save the World.” Technology Review, 17 Feb. 2020, www.technologyreview.com/2020/02/17/844721/ai-openai-moonshot-elon-musk-sam-altman-greg-brockman-messy-secretive-reality/. Accessed 8 Aug. 2023.

Hughes, Alex. “ChatGPT: Everything You Need to Know about OpenAI's GPT-4 Tool.” BBC Science Focus Magazine, 30 June 2023, www.sciencefocus.com/future-technology/gpt-3. Accessed 8 Aug. 2023.

Kay, Grace, and Sarah Jackson. "The History of OpenAI, from the Early Days with Elon Musk to the ChatGPT Maker Being Put on Blast by Scarlett Johansson." Business Insider, 22 May 2024, www.businessinsider.com/history-of-openai-company-chatgpt-elon-musk-founded-2022-12. Accessed 24 May 2024.

Metz, Cade. “OpenAI Plans to Up the Ante in Tech’s A.I. Race.” The New York Times, 14 Mar. 2023, www.nytimes.com/2023/03/14/technology/openai-gpt4-chatgpt.html. Accessed 8 Aug. 2023.

Metz, Cade, and Tripp Mickle. "Open AI Completes Deal That Values the Company at $80 Billion." The New York Times, 16 Feb. 2024, www.nytimes.com/2024/02/16/technology/openai-artificial-intelligence-deal-valuation. Accessed 24 May 2024.

Teo, Kai Xiang. “Adding One Line of Code Can Now Prevent OpenAI from Accessing a Website’s Data to Train ChatGPT.” Business Insider, 8 Aug. 2023, www.businessinsider.com/openai-code-how-to-block-using-website-train-chatgpt-2023-8. Accessed 8 Aug. 2023.