OpenAI Agrees in Principle to Sam Altman's Return as CEO, But this is Far From Over
What happened and what to expect
OpenAI announced on X that it had “reached an agreement in principle” to bring back Sam Altman as CEO. It also said in the 47-word post that the new board composition will include former Salesforce co-CEO Bret Taylor, who will serve as “Chair,” former U.S. Treasury Secretary Larry Summers, and existing board member Adam D’Angelo. Neither Altman nor Brockman will return to their board seats, which The Information reported is part of the compromise deal.
Notably, Ilya Sutskever, the company’s first technical leader and the co-founder who told Altman he was fired, also will not be on the newly constituted board. That means no original OpenAI co-founder will have board authority. The only board member with some tenure at the company will be Adam D’Angelo, who appears to have a significant conflict of interest overseeing a rival service to OpenAI’s ChatGPT.
A First Step
Sam Altman took to X late last night to confirm the news. He said he believes returning to his role along with a new board composition and Microsoft CEO Satya Nadella’s blessing is the “best path for me and the team.”
Nadella posted on X one minute later that he was encouraged by the developments. He also said, “We believe this is a first essential step on a path to more stable, well-informed, and effective governance.” The point here is that OpenAI’s work is far from over in resolving the crises that nearly tanked the entire enterprise. The tumult demonstrated its governance structure’s weakness.
Beyond governance, OpenAI may have permanently undermined the trust of its partners, developers, and customers. Bakz T. Future (AKA, Bakz Awan) became a developer, prolific user, and commentator on OpenAI in 2020. He issued a statement this morning on YouTube saying:
I would like to say here, on the record, that I feel I've put my trust in a bunch of teenagers. I feel like I actually believed that you all would be able to, through your organization, through your governance, through your structure, be able to handle the stakes and pressure of something like AGI or artificial super intelligence. After the unprofessionalism, after the immaturity, after the lack of communication that I saw this weekend, I highly question if OpenAI is capable of handling anything even close to AGI…
I hope OpenAI hits the ground running now that Sam is back immediately addressing this gap in public trust, reiterating their commitment to their mission and the changes they're going to make…This whole thing is making me rethink Cohere. It's making me rethink open source. It's making me rethink all these things. There is a counterargument here. Maybe a lot of the points these open-source people were making, such as putting our trust in a small handful of Silicon Valley companies is a misguided idea. In the beginning, my thinking was to give these companies and these incredible individuals the benefit of the doubt, but what we're seeing…this again may be just too much pressure for any small group of individuals.
Bakz took the time to share the thoughts of many people who have long been committed to OpenAI. You can hear more about these concerns from Leslie Pound and Bram Adams, which were expressed during Synthedia’s OpenAI developer livestream yesterday.
Re-establishing OpenAI’s team and its governance model may be the first step. However, that must be followed by re-establishing trust with OpenAI’s key constituencies. Some of that trust has been permanently lost. Over the past weekend, over 100 OpenAI customers reportedly reached out to Anthropic to inquire about evaluating its software. Other leading large language model (LLM) providers also experienced an influx of requests. The search for alternatives and mitigating vendor concentration risk is underway.
Weighing the Options
Beyond this, many developers began considering other vendor options during the period of uncertainty. Those options include open-source alternatives that carry a lower risk of organizational disruption. Everyone knew intellectually that they should have a backup plan and not be wholly reliant on a single generative AI foundation model vendor. What has changed is the urgency to assemble those alternative options has accelerated. OpenAI was creating a stranglehold on the industry.
A July 2023 analysis by Replit found that “160,000 of these [AI] projects were created in Q2 ‘23. That’s ~80% QoQ growth, and it is +34x YoY. We continue to see these numbers accelerate. The majority of these projects are using OpenAI. When we compare providers, OpenAI dominates >80% of distinct AI projects on Replit.
High awareness and confidence in OpenAI led to most projects initially building on its foundation models. Many companies used the initial proof of concept experiment with OpenAI models as justification to expand their solutions and begin the process of developing production-grade generative AI capabilities. While that typically involved considering other models and comparing them in terms of cost, latency, and quality, OpenAI was already in-house. Every other foundation model provider was and is fighting to become the OpenAI alternative.
That situation will remain the case next week. However, other models just became more attractive. The largest organizations will either be inclined to head toward open-source, knowing they have full control in exchange for more complexity and potentially higher cost, or establish options to switch models in case one provider becomes compromised. This may be a very good outcome for the industry as it will reduce the over-reliance on OpenAI. Still, it points out that OpenAI has undermined its market position and has a lot of work to do to re-establish it.
Broader Implications
Synthedia has published a full updated timeline of the OpenAI saga here. Check it out for more background information.
I expect four immediate fallout consequences from these events for business users of generative AI.
Establishing an AI Governance framework will become a higher priority.
Non-OpenAI models will receive increased interest, trial, and adoption.
The shift to considering OpenAI alternatives will lead to greater demand for benchmarking tools and services.
Many enterprise projects will slow their path to production as they work through risk mitigation, governance, and model comparison tasks.
I expect nothing to change from a consumer standpoint related to ChatGPT and GPTs. ChatGPT will remain the most popular generative AI-powered chat assistant. GPTs production and use will grow quickly after the marketplace is launched and they become more widely available to free users.
OpenAI still has many assets. GPT-4 is considered the most capable general-purpose generative AI model, OpenAI has the most generative AI enterprise customers by a wide margin, and ChatGPT has over 100 million weekly users. In addition, no other generative AI company commands the same level of interest across business, technology, developer, and government communities and in the general public.
Despite its attempts at self-immolation, OpenAI is still the leader in generative AI foundation models. It just doesn’t have as much of a lead today as it did a week ago.
Synthedia is a community-supported publication. Please check out our sponsor, Dabble Lab, an independent research and software development agency that is entirely automated with a mix of in-house AI tools and GitHub Copilot. The founder, Steve Tingiris, is also the author of Exploring GPT-3, the first developer training manual for building applications on GPT-3.
To become a sponsor of Synthedia and raise your visibility, please check out this post.
One outcome of this bit of drama is that OpenAI (and LLMs in general) can rebaseline at an entirely new elevated level of hype. Salivating investors want in on this rocket ship of a technology... Salesforce was offering to match the $10MM offers that Microsoft gave to employees as a signing bonus. The hype curve on this space just won't stop...