A Timeline of OpenAI’s Technology, Funding, and History
This excerpt is from my article ’48 Hours with OpenAI: Insights and Reflections — Amidst Intense Debates: Spreading Love and Understanding Through OpenAI’s Lens.’ To explore more insights and in-depth analysis, check out the full article here.
Too much has happened with OpenAI in the past 48 hours. I dived into OpenAI’s website and found much more information than I expected. I started with the About page, and from there, I read five impressive long articles: ‘The Vision for AGI’, ‘The Charter’, ‘The Structure of OpenAI’, ‘The Description of the Capped-Profit’, and ‘Why I’m Optimistic About Our Alignment Approach’ by Jan Leike.
I summarized the history of OpenAI’s capped-profit structure as follows, based on public information, as I gradually understood more about it.
I tried my best to gather information from diverse sources, filtered them with my understanding, and described it in a more objective way. I hope to find primary source information unless it was not clear or I did not have a paid subscription to the source.
2015:
OpenAI was founded by Sam Altman, Greg Brockman, Reid Hoffman, Jessica Livingston, Peter Thiel, Elon Musk, Amazon Web Services (AWS), Infosys, and YC Research announced with a $1 billion pledge from investors.
2016:
In April 2016, OpenAI launched its first products, OpenAI Gym, an open-source toolkit for reinforcement learning. In Dec 2016, OpenAI released Universe, a platform for AI agent training.
2017:
OpenAI allocated $7.9 million, a quarter of its functional expenses, specifically for cloud computing.
For comparison, DeepMind’s total expenses in 2017 amounted to $442 million.
2018:
OpenAI released GPT-1, a groundbreaking language model with 117 million parameters capable of generating diverse texts.
In June, OpenAI rented 128,000 CPUs and 256 GPUs from Google for several weeks to train its Dota 2 bots OpenAI Five.
OpenAI entered a multi-year partnership with Amazon Web Services (AWS) for cloud services, granting AWS exclusive rights to market and sell OpenAI’s technologies.
OpenAI introduced Dactyl, a tool that uses machine learning to train a robotic hand for manipulating objects. Dactyl learns entirely through simulation using the reinforcement learning algorithm and training code. It has motion tracking and RGB cameras that manipulate arbitrary objects
Elon Musk resigned from OpenAI’s board due to potential conflicts of interest with Tesla’s AI development but remained as a donor.
2019:
In February, OpenAI opted not to release GPT-2, a new AI fake text generator, due to concerns about potential misuse.
In March 2019, Sam Altman left Y Combinator to focus on his work at OpenAI, and became the CEO of OpenAI.
On March 11, OpenAI introduced OpenAI LP as a ‘capped-profit’ entity. This structure enhances investment in computing and talent and limits investor returns to 100 times their investment, all under the oversight of a nonprofit board.
In July 2019, OpenAI partnered with Microsoft, backed by a $1 billion investment, to advance Azure AI supercomputing and AGI development.
On November 5, 2019, OpenAI released the final version of GPT-2, its largest model with 1.5 billion parameters, along with tools to detect GPT-2-generated outputs. Although larger models had emerged since August, OpenAI worked with researchers to assess misuse risks and found ‘no strong evidence of misuse so far.’
2020:
OpenAI was headquartered in San Francisco’s Mission District.
In June 2020, OpenAI announced GPT-3, a language model trained on trillions of words from the Internet.
Jan 23 2020 OpenAI published the paper ‘Scaling Laws for Neural Language Models’ on arXiv. The results were similar to another paper ‘A Constructive Prediction of the Generalization error across scales’ published at ICLR 2020.
The partnership with Microsoft’s $1 billion investment in OpenAI continued to be significant throughout 2020.
2021
On January 5, OpenAI introduced DALL-E, a deep-learning model capable of generating digital images from natural language descriptions.
OpenAI launched Codex, an AI-powered programming tool adept at twelve programming languages including Python and Java, and capable of analyzing code blocks, anticipating the next line, and generating code. Codex is the driving technology behind GitHub CoPilot.
May 26, 2021, OpenAI has raised a total of $100M in a single venture fund, OpenAI Startup Fund to support its ongoing development and initiatives.
In July 2021, Microsoft made an additional investment of $2 billion in OpenAI.
During 2021 and early 2022, several renowned venture capital firms, including Bedrock Capital, Sequoia Capital, Andreessen Horowitz, Tiger Global Management, and Matthew Brown Companies, invested in OpenAI, valuing it between $14 billion and $20 billion.
2022
In December 2022, OpenAI launched the preview of ChatGPT based on GPT-3.5.
OpenAI reported a staggering loss of $540 million in 2022, generating only $28 million in revenue.
2023
OpenAI Global LLC was in discussions for funding that would value the company at $29 billion
In January 2023, Microsoft announced a multi-year, $10 billion investment in OpenAI Global LLC. Rumors of this deal suggested that Microsoft may receive 75% of OpenAI’s profits until it secures its investment return, and a 49% stake in the company.
On March 14, 2023, OpenAI released GPT-4, both as an API and as a feature of ChatGPT Plus
In May 2023, Sam Altman, Greg Brockman and Ilya Sutskever posted recommendations for the governance of superintelligence, highlighting the need for international oversight and increased safety research for advanced AI systems.
On Aug 16, 2023 OpenAI acquired Global Illumination for $100M
On September 6, 2023, OpenAI announced OpenAI’s first developer conference in San Francisco.
On Sep 16, 2023, OpenAI invested in KudoAI. This investment was valued at $98K. On November 6, 2023, OpenAI held the first developer conference, and launched GPT for users to create customized ChatGPT
On November 6, OpenAI hosted its first developer conference in-person in San Francisco and livestream the keynotes. Recordings of the keynote and breakout sessions are available to the public one week after the event.
Many key announcements were made: GPT-4 Turbo, Users Building Own GPTs, GPT Store for User-Created AI Bots, New Assistants API, DALL-E 3 API, New Text-to-Speech APIs, Copyright Shield Program, additional updates included a program for companies to build custom models with OpenAI’s help, the removal of the model picker in ChatGPT for simplicity, the launch of Whisper large v3 (an open-sourced speech recognition model), and doubling the token limit rate for paying GPT-4 customers
On November 17, 2023 OpenAI announced the leadership transition that ‘Sam Altman will depart as CEO and leave the board of directors. Mira Murati, the company’s chief technology officer, will serve as interim CEO, effective immediately’.
❤️I think it’s important to note that OpenAI has been making efforts to broaden its talent pool. The efficacy of these efforts may be open for discussion❤️
OpenAI initiated its remote Scholars program for underrepresented minorities in 2018. The OpenAI Scholars program offers mentorship and promotes diversity, attracting scholars from fields like computer science and neuroscience. The final project demonstrations by OpenAI scholars each year are remarkable.
On November 30, 2021, OpenAI announced the OpenAI Residency, a six-month program, which offers a pathway to a full-time role at OpenAI for researchers and engineers who don’t currently focus on artificial intelligence.
Check out more final projects here:
https://openai.com/blog/openai-scholars-2018-final-projects
https://openai.com/blog/openai-scholars-2019-final-projects
https://openai.com/blog/openai-scholars-2020-final-projects
https://openai.com/blog/openai-scholars-2021-final-projects