13 is a beautiful number. With this number, we wrap up our series on Generative AI Unicorns. So far, we've chronicled OpenAI, Anthropic, Inflection, Hugging Face, Cohere, Lightricks, Runway, Jasper, Replit, Character AI, AI21 Labs, Mistral AI, and Stability AI. Most likely, we will update this list and share more investigations as new unicorns with particularly interesting stories emerge, but for now – this episode is the last in the series. And it is about... OpenAI again: An Update.
We published our first article about OpenAI 8 (!) months ago when we had just a little over 1,000 readers. Now, with over 46,000 readers, we want to revisit their story, adding a few important updates about this titan of the GenAI world.
Enjoy the ride.
(FYI, stay tuned for the next series: it will be about AI Infrastructure Unicorns. They provide the hardware, software, and services necessary for generative AI companies but even if GenAI will someday become extinct, these infrastructure builders won’t stay without the job as they serve a much bigger industry of AI/ML models in general. Think Databricks, CoreWeave, Scale AI, etc. – for each of them, we will find a unique angle to cover!)
In the first part: OpenAI Chronicle: What Drives ChatGPT Creators, we covered:
Founder’s intentions and internal friction
Sam Altman’s vision
Money situation and current investments
GPT development history and other interesting products
Safety & alignment
Urge for regulations
Bonus: What and Who is Open AI
It was published on June 07, 2023. So many things happened since then! Dive in →
At the end of June, CEO Sam Altman returned from a massive global tour. He traveled to 25 cities across 6 continents to speak with users, developers, policymakers, and the public, and hear about each community’s priorities for AI development and deployment. Nothing really interesting, but after the trip, the company summarized: Insights from global conversations.
In August, we talked to OpenAI General Counsel, Che Chang. We discussed how the OpenAI legal team uses ChatGPT, how they navigate legal uncertainty around new technology, and how they operate across different jurisdictions. Che also offered guidance for enterprises on navigating risks and liability when working with GenAI and provided insights into the future.
In general, OpenAI was moving at cosmic speed, delivering updates one after another. It all led to an exciting DevDay at the beginning of November, where OpenAI, among other things, introduced GPTs. GPTs allow for the customization of ChatGPT for specific uses, enabling the creation and sharing of tailored AI versions without coding, for tasks ranging from learning games to designing stickers, thereby enhancing daily life and work productivity. They also promised to open a sort of marketplace for them in the nearest future.
On November 9th, they announced OpenAI Data Partnerships: “We’ll work together with organizations to produce public and private datasets for training AI models.” “To ultimately make AGI that is safe and beneficial to all of humanity, we’d like AI models to deeply understand all subject matters, industries, cultures, and languages, which requires as broad a training dataset as possible.”
And then – like thunder out of a clear sky – in mid-November Sam Altman was dismissed from CEO position.
November 22, 2023 – Board changes
OpenAI became a king of drama as the company faced the board meltdown and recovery in just two weeks. It all started on Nov 17, 2023, Friday, with a statement from the OpenAI board dismissing Altman from the CEO position, removing Greg Brockman from the board, and placing Mira Murati, a CTO, as an interim CEO. We covered this flow of events in one of our FODs, highly recommended to read.
There is also a pretty interesting chronology of tweets to follow from Sam Altman. Just to summarize it for you:
Sam wrote about the “transformative” experience at OpenAI and “his love to OpenAI” in several tweets. There was even an “OpenAI is nothing without its people” flash mob all over the X (Twitter) with people posting this phrase. Employees threatened to resign en masse and go work for Microsoft, which is backing OpenAI with billions of dollars.
After 12 intense days, during which the top executives of OpenAI had barely any sleep, the company announced that Sam Altman returned as CEO, Mira Murati as CTO, and Greg Brockman returned as President. But that’s not all. The company’s initial board experienced a significant shakeup, appointing Bret Taylor as the new chair and welcoming two new members: Larry Summers and Adam D’Angelo. Adding to the drama, Altman revealed that Ilya Sutskever, a chief scientist and co-founder, would no longer serve on the board.
“I love and respect Ilya, I think he's a guiding light of the field and a gem of a human being. I harbor zero ill will towards him. While Ilya will no longer serve on the board, we hope to continue our working relationship and are discussing how he can continue his work at OpenAI.”
If you don’t have time to read the full announcement, here is our outline:
Wojciech Zaremba (Co-Founder of OpenAI) reaction on Twitter to these changes:
What was the coup about? According to the New York Times: “The coup was led by Ilya Sutskever, OpenAI’s chief scientist, who had butted heads with Altman. Sutskever wants the company to prioritize safety and was worried that Altman was more focused on growth.”
Earlier, Sutskever was one of the two authors of the Superalignment team announcement “to steer or control a potentially superintelligent AI, and prevent it from going rogue.”
It was also noted that all female members were removed from the board. Although the board members pledged to focus on diversity, as of today – February 2 – the board still consists of only three male members.
Enough drama, let’s see what was happening with →
ChatGPT updates
ChatGPT can now see, hear, and speak with New voice and image capabilities. But also this title anthropomorphizes AI
Introduction of DALL-E 3, it is now only available via ChatGPT and there is no separate interface for it.
“DALL·E 3 is the culmination of several research advancements, both from within and outside of OpenAI. Compared to its predecessor, DALL·E 3 generates images that are not only more visually striking but also crisper in detail.” The OpenAI researchers published a paper, where they presented an approach to enhance text-to-image models' prompt-following abilities by training on refined, synthetic image captions. Existing models often struggle with detailed prompts, losing precision or distorting meanings. The authors hypothesized that the root cause lies in the training data's poor-quality captions. By developing a specialized image captioner to generate highly descriptive captions for dataset images, they retrained text-to-image models, resulting in significantly improved prompt adherence.
And, as we mentioned earlier, OpenAI shipped customizable GPTs.
Product and research updates
September 25, 2023 – GPT-4V(ision) system card
October 3, 2023 – DALL·E 3 system card
November 2023 – besides GPTs, new models and developer products were announced at DevDay. People were posting about it all over Twitter. GPT-4 Turbo with 128K context and lower prices, the new Assistants API, GPT-4 Turbo with Vision, DALL·E 3 API, and more.
Acquisitions and partnerships
OpenAI has acquired the team at Global Illumination, a company that has been leveraging AI to build creative tools, infrastructure, and digital experiences.
Partnership with Axel Springer became the first global publishing house to strike a licensing deal with OpenAI. They claim “to deepen beneficial use of AI in journalism.” This collaboration aims to intertwine AI technologies with journalistic content, potentially revolutionizing how we access and engage with news. By incorporating Axel Springer's reputable media brands into ChatGPT, the partnership promises enhanced content experiences for users and new revenue avenues for journalism. Will it redefine the landscape of journalism through AI? It surely protects OpenAI from being sued (like the New York Times did just recently – good analysis of this story on Twitter).
Safety & alignment initiatives
In a proactive move, the Biden-Harris Administration has garnered voluntary commitments from seven leading AI firms, including Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI, aiming to pave the way for safe, secure, and transparent AI development. These commitments revolve around ensuring AI product safety prior to public release, emphasizing security-first systems, and earning public trust through transparency and responsibility. How might these steps, coupled with an Executive Order and bipartisan legislation in the works, shape the future of AI innovation and its societal impact?
OpenAI detailed its commitments here, concentrating on Safety, Security, and Trust.
Anthropic, Google, Microsoft, and OpenAI established the Frontier Model Forum, aimed at fostering the safe and responsible development of advanced AI models. With ambitions to leverage AI for societal good, the Forum appears poised to shape the future of AI development. We wonder how its efforts, including the formation of an Advisory Board and the invitation for broader participation, will influence the trajectory of AI safety and responsibility. They also announced a new $10 million AI Safety Fund, which doesn’t seem a lot.
OpenAI announced the Red Teaming Network. Domain experts were encouraged to join the Network to improve the safety of OpenAI’s models.
They also announced the Preparedness team. Led by Aleksander Madry, the team aims to evaluate, monitor, and mitigate potential dangers, including misuse and catastrophic capabilities. Additionally, OpenAI is launching an AI Preparedness Challenge to encourage innovative solutions for misuse prevention, offering rewards and potentially recruiting talent from participants.
2024 OpenAI started with an explanatory blog about how they are going to approach worldwide elections. Efforts include preventing misuse of AI tools, enhancing transparency regarding AI-generated content, and facilitating access to accurate voting information. Collaborating with various stakeholders, OpenAI aims to ensure its technologies uphold election integrity through measures like red teaming for safety, policy refinements to combat abuse, and partnerships for disseminating authoritative voting details.
Military collaboration
Time reported that “OpenAI is working with the Pentagon on a number of projects including cybersecurity capabilities, a departure from the startup’s earlier ban on providing its artificial intelligence to militaries.”
The Intercept wrote that “OpenAI quietly deletes ban on 'military' use from terms of service.”
Education projects
Khan Academy, Canva, Duolingo, edX became some of the first educational partners to build on top of OpenAI technology.
OpenAI released a guide for teachers using ChatGPT in their classroom.
Arizona State University announced a new collaboration with OpenAI to “bring the advanced capabilities of ChatGPT Enterprise into higher education, setting a new precedent for how universities enhance learning, creativity, and student outcomes.”
OpenAI partnered with Common Sense Media to educate teens on safe AI use. Common Sense Media, focusing on technology safety for kids, will develop AI ratings and review systems. The partnership seeks to create AI guidelines and "family-friendly" GPT models.
These were just a few highlights of what we think is important that has happened to OpenAI since we published OpenAI Chronicle. Please read the Chronicle itself; it's an insightful investigation into how the founders of OpenAI think and what they try (dream?) to achieve: What is Sam Altman's vision, how OpenAI makes money, what caused ChatGPT success, and the urge for AI regulations.
You can also check the other 12 Chronicles about GenAI Unicorns:
Stability AI Chronicle: Why It Can Fail in 2024. Investigating the Thin Line Between Genuine Innovation and Strategic Exaggeration in AI.
Mistral AI's Bold Journey. From Paris to Global Stage: The Unconventional Rise of a French AI Unicorn and Its Open-Source Revolution.
Jasper AI: A Dilemma of a Thin Wrapper. GPT's Blessing and Curse on The Unicorn Turned Underdog.
Anthropic Chronicle: Unveiling the Secrets of OpenAI's Rival. What vision drives it, how is it connected to effective altruism, and what is so special about Claude?
Inflection AI: The Story of the Unforeseen and Secretive Unicorn with a $4 Billion Valuation. Why Reid Hoffman and Mustafa Suleyman, veterans of tech and AI, are so passionate about Inflection Pi, their business strategy and thoughts about AI risks.
Democratizing AI: The Hugging Face Ethos of Accessible Machine Learning. How 3 founders turned their passion for sharing knowledge into a thriving ML community.
Cohere AI Chronicle – Challenging Tech Giants With AI For All. Bringing Google-Quality AI to Everyone While Addressing Concerns About LLMs Biases.
AI21 Labs: Being in Amazon's Bedrock without any investment from Amazon. How AI21 Labs Maintains Independence and Innovation While Collaborating with Amazon, Distinguishing Itself from Industry Giants Like OpenAI.
Character AI's Journey: From Not Needed to Highly Valued. Challenging Conventions: How Character AI is Carving Its Niche Amidst Tech Titans (and Can It?)
Runway Chronicles: Melding Art and AI in a Transient Tech Terrain. Watch Runway's evolution, learn the tech behind Stable Diffusion, as well as legal challenges, financials, and the company's mission to democratize generative AI.
Lightricks: The Oldest Generative AI Unicorn. How to leverage AI for creator economy.
Stay tuned for the next AI Infrastructure Unicorns Series: we are aiming at Databricks!
