In this monthly roundup, we highlight the most noteworthy AI news stories from November:
OpenAI’s Leadership Transition
On November 17, OpenAI made an announcement regarding a leadership transition. In this change, Sam Altman departed from his CEO role and left the board of directors. The board has appointed Mira Murati, OpenAI’s CTO, as the interim CEO. Altman’s departure was prompted by a thorough review process revealing inconsistencies in his communications with the board, hindering its ability to fulfill its responsibilities. As part of this transition, Greg Brockman, the chairman of the board, also stepped down from his position.
Following the announcement, over 95% of OpenAI’s staff, including interim CEO Mira Murati, signed an open letter advocating for Altman’s return. Meanwhile, Altman had already been hired by Microsoft, OpenAI’s primary investor. However, an agreement in principle was rapidly reached, leading to Altman’s return as OpenAI’s CEO, accompanied by Greg Brockman as Director and a reshaped initial board consisting of Bret Taylor as Chair, alongside Larry Summers and Adam D’Angelo. Following Altman’s return, Microsoft also got a non-voting observer seat on the board.
Between the lines: This unclear and unprecedented situation garnered widespread attention and discussion, sparking speculation and negative connotations among many within and beyond the tech industry. This case will probably find its way into films or books one day, shedding light on the details. Yet, the true reasons behind the transition remain known only to the players involved.
OpenAI’s New Features
A few days before the leadership transition OpenAI unveiled a slew of new features at its inaugural developer conference. Firstly, the company is set to launch the GPT Store later this month — a platform empowering users to share and sell their custom GPT bots. Notably, Sam Altman announced a commitment to reward those who build the most useful and widely used GPTs with a share of the company’s revenue.
Additionally, OpenAI introduced the “Copyright Shield,” available to ChatGPT Enterprise and API users. The program is about to defend customers and cover the costs incurred in the face of legal claims related to copyright infringement.
The company also unleashed major enhancements to its language models, GPT-4 and GPT-3.5, offering updated knowledge bases and a significantly expanded context window. The introduction of GPT-4 Turbo, available via API preview, marked a notable leap forward, having been trained with information up to April 2023.
OpenAI’s commitment to democratizing AI takes center stage with the launch of a platform allowing ChatGPT Plus subscribers to create custom versions of ChatGPT without coding. The platform enables the development of custom AI agents for specific tasks, which will also be available through the GPT Store, with creators earning compensation based on usage. The platform also grants enterprise customers the ability to create internal-only GPTs.
Why it matters: The release of OpenAI’s GPT Store marks a significant milestone, drawing parallels to the impact of the App Store’s launch. Serving as the primary provider of the algorithm, OpenAI facilitates the emergence of numerous bots, allowing users to generate income with minimal or no team involvement. Ben Tossell’s post, featuring an AI idea brought to life by a reader in just one hour, serves as a compelling example of the GPT’s potential.
Elon Musk Unveils AI Chatbot “Grok” from xAI Startup
Elon Musk has introduced a new chatbot named “Grok” from his xAI startup, emphasizing its commitment to making AI “useful to people of all backgrounds and political views.” Described as a “very early beta product” with a sense of humor and a “rebellious streak,” Grok is about to answer provocative questions that many other AI systems reject. Positioned as more than just a chatbot, Grok also stands out with its real-time knowledge of the world scrapped via the 𝕏 platform. Beyond responses based on real-time data, Grok AI will also have add-on features like a “fun” mode, multi-tasking, shareable chats, and feedback for conversations.
The context behind: The AI startup, introduced earlier this year, boasts a team with alumni of DeepMind, OpenAI, Google, and Microsoft. Musk, having previously banned algorithm training using data from X (formerly Twitter), has now released the chatbot based on that data. Musk seems to be setting it up to compete with OpenAI, a company he used to invest in.
Stability AI Releases
On November 21, Stability AI marked a significant milestone with the release of Stable Video Diffusion, their pioneering foundation model for generative video, built upon the success of the image model Stable Diffusion. The release includes open access to the code for Stable Video Diffusion on their GitHub repository, with the essential weights for local model execution hosted on their Hugging Face page. The model is easily adaptable to various downstream applications, including multi-view synthesis from a single image, with the option for fine-tuning on multi-view datasets. Stability AI also plans to expand its features with a Text-To-Video interface, showcasing the practical applications of Stable Video Diffusion in diverse sectors such as advertising, education, entertainment, and beyond.
Why it matters: Considering that Stability AI is an open source, the success of their video generation model could significantly influence the landscape with a large number of videos potentially being created through models based on Stable Video Diffusion. Drawing parallels with our study on AI-generated images, where Stable Diffusion played a major role, it’s highly possible that a similar trend may emerge in the realm of video generation.
Apart from the Stable Video Diffusion release, Stability AI unveils SDXL Turbo — a real-time text-to-image generation model. Accessible on Hugging Face under a non-commercial research license, SDXL Turbo outperforms multi-step models in blind tests, offering exceptional efficiency without compromising image quality.
The context behind: Earlier this year, Stability AI made a significant move by acquiring Init ML, the creators of the Clipdrop application. Clipdrop was the globally popular imaging tool known also for the AR algorithm capable of grabbing objects from the real world and placing them into desktop apps. Initially, the practical application of this AR technology remained uncertain, but it turned out to be highly utilitarian, especially in the field of design. Clipdrop laid the foundation, leading to the development of SDXL Turbo.
Miscellaneous
The New Yorker Cover
Christoph Niemann’s Illustration on the New Yorker November cover provokes thought on AI in Art. Addressing concerns about AI replacing creators, Niemann sees the current AI-generated art as superficial and emphasizes the importance of human intent in artistic expression.
Authentic and Hallucinate Named Words of the Year
Merriam-Webster has declared “authentic” as its Word of the Year for 2023, citing a substantial surge in online searches throughout the year. Merriam-Webster highlights the challenge in defining authenticity, which sparked widespread interest amid the growing blur between ‘real’ and ‘fake,’ accentuated by the AI boom.
The Cambridge Dictionary named “hallucinate” as its Word of the Year for 2023, reflecting the profound impact of GenAI on language. This decision follows the surge around AI, leading lexicographers to introduce new terms like LLM, GenAI, and GPT into the dictionary. Existing words, including “hallucinate,” have acquired additional meanings in the context of AI, illustrating the profound impact of this technology on linguistic evolution.
Happy ChatGPT Day
On November 30, ChatGPT marked its first birthday, and social media was buzzing with cute and grateful posts: