From Dream to Reality: The Inspiring Story of OpenAI's Journey from Scratch
11 min read
Disclaimer: As much as I would have loved to start this story-telling experience with something along the lines of "like your attention span, this story is short", OpenAI has come a long way. As such, this is quite the story. So sit tight, get your attention span in order, fasten your seatbelts, and get ready for a tale that will leave you inspired (and breathless).
That Wednesday afternoon in November 2022 was no different from a typical one in San Francisco. The fog was rolling in, the cable cars were ringing their bells, and the tech industry was buzzing with excitement. In the heart of this vibrant city, this group of brilliant minds who had come together to embark on a journey had changed the world of artificial intelligence. They looked around the room and knew this was the result of their toil and efforts for almost seven years prior. The journey to this point had been long and arduous. They had faced countless obstacles and setbacks but they knew they had made a breakthrough that would redefine the field of AI. As they looked out over the beautiful San Francisco skyline, they knew it was a special day, and that they were part of something special.
OpenAI had just released that artificial intelligence chatbot that can pass the bar exam - ChatGPT. You see, a few times in every generation, a product comes along that hurtles a technology out of the dimly-lit basements of engineering departments and the late-night experimentation of research labs to something your aunt who works as a sales representative at some random paper company in Scranton, Pennsylvania or that your distant cousin with the bowl-shaped haircut and eccentric clothing choices who works as an aerospace engineer at Caltech both get to use. This was one of those times, and this is the story of how OpenAI made that possible.
Trailblazers: A look back at the birth and formative years
- The First Day at OpenAI (Source)
OpenAI was founded in December 2015 as an artificial intelligence research company to develop advanced AI systems that could benefit humanity as a whole. OpenAI's goal at the time - to build AGI (i.e. machines or systems that can perform any intellectual task that a human can, in a general and flexible way) seemed like a very ambitious goal. Now, OpenAI was not the first organization to openly announce its pursuit of AGI. DeepMind had already declared its intentions five years earlier and had been bought by Google in 2014. However, OpenAI appeared to stand out from the rest, especially with how much initial funding they boasted of. The initial funding was surprising, with private investors putting up a staggering $1 billion. Started by a group of prominent figures in the technology industry, OpenAI's major players at the time included Sam Altman [now CEO of OpenAI and then Chairman of Y-Combinator], Elon Musk [CEO of Tesla, SpaceX and Twitter], Reid Hoffman [Former executive vice president of PayPal and Founder of LinkedIn], Jessica Livington [one of the founding partners at Y-Combinator], Ilya Sutskever [a research scientist at Google Brain at the time] and Peter Thiel [Co-founder of PayPal and Palantir Technologies], all of whom collectively pledged $1 billion.
Although these six players were the pioneer investors, paramount to OpenAI’s early growth from an engineering point of view was Greg Brockman who was the CTO at Stripe. During its first year of operation in 2016, OpenAI focused primarily on building its research team. The company hired several prominent researchers and engineers in the field of AI, including Ian Goodfellow, who created the popular machine learning technique known as generative adversarial networks (GANs).
With the research team formed, the company's focus was now on the creation of artificial intelligence geared toward video games and other related applications. This resulted in the development of its initial tools, such as the open-source toolkit for reinforcement learning known as OpenAI Gym and Universe. This tool essentially served as a platform for training AI agents. From 2017 to 2019, the company will go on to shift its attention toward more generalized AI research and development. During this period, they would release a research paper titled "Improving Language Understanding by Generative Pre-Training" which introduced the concept of a Generative Pre-trained Transformer (GPT). Yes, GPT. More on that later on.
"I would not bet against him" - Sam Altman, 2023
Elon Musk later left his board seat citing “a potential future conflict of interest” with his role as CEO of Tesla due to Tesla’s AI development for self-driving cars but remained a donor. There are rumors that this happened as an aftermath of his poaching of one of the top scientists at OpenAI, a Slovak-Canadian computer scientist in the person of Andrey Karpathy [who joined OpenAI back last month claiming that he is very inspired by the impact of their work and that he has personally benefited greatly from it] to become the head of computer vision of Tesla Autopilot.
Fast forward to January 2023 when Sam Altman, the CEO of OpenAI spoke at a StrictlyVC event about the latest at OpenAI. He answered questions about how his other investments fit into larger themes that he expects to play out shortly, and when asked about Elon Musk and his work at Twitter, Sam said although he would do a number of things differently, he would not bet against Elon.
Note: Last night, another board member, Reid Hoffman announced that he has decided to take a step back from the OpenAI board. Reid who is a seasoned entrepreneur and a member of the board at Microsoft and OpenAI claims that he will be investing in companies that will use the OpenAI APIs and believe his position on the board of OpenAI will potentially look like it is leading to differential economic.
Bringing Imagination to Life: The GPTs
In my humble opinion, what makes OpenAI's work stand out is how they seek to bring imagination to life like we have never seen before. Nothing quite encapsulates how different their work (with their image-from-text generator DALL-E in this case) is as much as this excerpt from a blog post I wrote seven months ago.
OpenAI's first step towards this dates as far back as June 2018 when they developed GPT-1 (Generative Pre-trained Transformer 1). It would be the first model in the GPT series of large language models (LLMs) that are designed to generate human-like text by predicting the next word in a sequence of text. Nine months later, the second version, GPT-2 was launched. Typically in engineering, second versions tend to incorporate improvements and new features that address the limitations and shortcomings of the first one and this was no different. While GPT-1 was trained on 7000 unpublished books, GPT-2 was trained on 8 million documents that were scraped from Reddit. Talk about an upgrade!
However, that was not all - not even close. In May 2020, GPT-3 whose use would soon change everything for the company was released. It was trained on 570Gb of plain text and unlike the other GPT series, it was trained for more than one epoch. For context, an epoch is an iteration over the entire data provided and GPT-3 was iterated over the entire data more than once. GPT-3 will form the basis for GitHub Copilot, DALL-E and of course, ChatGPT.
It is important to note that in November 2019, OpenAI made the GPT-2 language model available for public use and released the model's code and pre-trained models. This move was aimed at encouraging research and development in the field of natural language processing and promoting responsible AI development practices. Fair to say 2019 was a busy year for the company because also in that year, they formed a partnership with Microsoft, which included a $1 billion investment from Microsoft to support OpenAI's research and development efforts.
In 2021, OpenAI launched a 12-billion parameter version of GPT-3 trained to generate images from text descriptions called DALL-E. The application combines the wonderful craziness of Salvador Dalí with the make-life-easy attribute of the doe-eyed Disney robot, WALL-E from the Pixar film of the same name.
The dream becomes reality: Research papers to one million users in five days
Facebook, ten months. Twitter, two years and four months. Instagram, two and a half months. Pinterest, nine months. TikTok, six months. Snapchat, four months. WhatsApp, nine months. ChatGPT, five days.
That's how long it took these companies to get their first one million users. So now, when those great individuals looked over the San Francisco skyline, boy, do you understand what it meant and the magnitude of what they had created! In hindsight, it has become clear that the release of ChatGPT was the initiation of a domino effect of great moves for OpenAI.
Microsoft went on to extend its partnership with OpenAI with a “multi-year, multi-billion-dollar” investment. Following this investment, Microsoft now owns 49% of OpenAI. They have integrated ChatGPT with MS Teams to make it more intelligent and even rolled out ChatGPT-enabled New Bing Search Engine. Also, OpenAI released a new subscription plan for ChatGPT available for $20/month that will grant users general access to ChatGPT even during peak, faster response times and priority access to new features. They also formed an alliance with Bain and Company in a partnership that would combine OpenAI’s industry-leading artificial intelligence tools and platforms with Bain’s strategic guidance and digital implementation capabilities. Two days ago, OpenAI made ChatGPT and Whisper APIs available in a bid to allow developers to integrate ChatGPT and Whisper models into their apps.
The New Kid in the Block: Anthropic, Claude and everything in between
Amazon and Walmart. Coca-Cola and PepsiCo. Boeing and Airbus. Nike and Adidas. Uber and Lyft. You know what, with the AI race, dare I add Google and Microsoft? It does not take a genius to know that the AI race between Microsoft and Google (and other tech companies) is ongoing and very intense. Both companies are investing heavily in research and development in a bid to outdo each other with innovation and market share. On a personal note, I find this beautiful to watch because this competition is driving rapid advancement in AI technology, which in turn is benefiting businesses and consumers alike as seen with the near-simultaneous announcement of Google’s Bard and Microsoft Bing Chat. I say the genesis of this race traces as far back as 2019 when in a deal that made them the “exclusive” provider of cloud computing services to OpenAI, Microsoft invested $1 Billion in OpenAI. Furthermore, as part of this deal, OpenAI would also license some of its techs to Microsoft to commercialize.
Also in 2019, OpenAI announced that it was changing its legal structure to a for-profit company stating that this change would allow it to attract more investment and pursue a broader range of research and development initiatives. Keep in mind that OpenAI was initially founded under the premise of we-want-to-advance-artificial-intelligence-in-a-safe-and-beneficial-way-for-humanity and that’s it. Fair to say that this “change of direction” did not sit well with some of the staff members at OpenAI. Citing directional differences in tandem with the company’s increasing commercial focus after its first deal with Microsoft in 2019, some staff members opted to leave OpenAI in 2021.
Daniela Amodei who was the Vice President of Safety and Policy at the time, Dario Amodei (siblings with Daniela) who served as the Vice President of research at the time, Tom Brown who was a senior member of technical staff and even led the engineering of GPT-3 and four others came together to form a startup of their own called Anthropic. As you probably guessed, Anthropic is an AI safety and research company, and their work is towards building reliable, interpretable and steerable AI systems. Now, picture this; members of staff felt dissatisfied with the direction of a firm, so they came together, quit the firm and started a new firm of their own together to effect the proper direction they feel the firm should have gone. That is not out of the ordinary now, is it? I reckon that it is not an uncommon scenario.
Here’s where it gets a little uncommon. In late 2022, Google (yes, Google) invested nearly $400 million in…Anthropic. In point of fact, the dynamic of their marriage is similar to the initial partnership between Microsoft and OpenAI - Anthropic, like OpenAI, would provide research expertise while Google, like Microsoft, will provide not only hundreds of millions of dollars in investment but also access to their colossal cloud platforms needed to train the compute-intensive AI models.
To be frank, whilst I find the seeming new turn of events amusing, it remains unclear how much the company would be commercially focused. Anthropic eventually created its own AI chatbot, called Claude, which operates through a messaging interface and provides users with detailed responses to their requests and inquiries, much like ChatGPT. Initially available in a closed beta via Slack integration, Claude can now be accessed through the Poe app developed by Quora. The app is currently available for iOS and will soon be available for Android.
I pride myself on being a great writer and storyteller. However, I would be doing this entire story a bit of disservice if I do not mention how much help ChatGPT proffered in making this story what it is. As we have seen since the launch of ChatGPT, AI will continue to transform the way writers (and everyone else who cares to join the party) work and help streamline the creative process. However, I would like to point out that I opine that whilst AI is undoubtedly a powerful tool for writers, it cannot replace them. While AI can certainly help with tasks like grammar and syntax checking, generating ideas, and even providing some basic content suggestions, writing is a deeply human endeavor, and it requires a level of empathy and insight that machines are currently unable to replicate.
But then, what if I told you the entirety of this blog post was written by ChatGPT? Well, I would be lying. My fingers are crossed that AI proves me wrong again. If you enjoyed reading this story-telling experience and have any thoughts or comments to share on AI replacing writers, I would love to hear from you! Please feel free to like, share, and leave your feedback below. Finally, I share my writings on Twitter if you enjoyed this article and want to see more.