How OpenAI - and Sam Altman - Fumbled the Biggest Tech Lead in History

And was this about economic forces, or hubris?

OpenAI is currently in the middle of experiencing one of the most colossal missed opportunities in modern technology. OpenAI’s incredible success came with the promise of transforming the nonprofit into the only developer of the largest language models. But by losing wave after wave of the most safety-oriented people in the company, Sam Altman’s untrustworthiness, coupled with a series of missteps, has instead left OpenAI struggling to maintain its once-enviable lead.

The organization, founded as a nonprofit with the goal of ensuring AI's benefits were widely distributed, was effectively alone in the LLM space through 2021. Until then, the secret sauce for these models, in terms of their construction, training, and infrastructure, was effectively monopolized by the company. (Similar to Deepmind’s contemporaneous lead in RL-based AI systems.) But OpenAI, which after the exit of Elon Musk was led by Sam Altman, moved to commercialize the product, starting with their creation of a for-profit subsidiary in 2019. The introduction of a for-profit entity within the nonprofit structure may have been necessary in order to keep pay levels competitive with other companies, but critics - both internal and external - argued it compromised the founding principles

Given that, it was unsurprising that OpenAI lost a dozen key employees, and its monopoly on the expertise building LLMs, in 2021, with the founding of Anthropic. And this was clearly and directly due to alienating Dario Amodei and others who were more concerned about safety than about speed or commercialization. To compete, they went to Google for their infrastructure needs - which happened, coincidentally or not, just as Deepmind quickly began to catch up with SoTA in LLMs.

But Altman’s decisions didn’t just squander their lead—they also contributed to even more internal unrest, and frustrated OpenAI’s original mission. He alienated board members and senior leadership, many of whom had originally championed the nonprofit’s vision. The fracturing of trust within OpenAI’s leadership led to the highly publicized firing of Altman. The commercial implications, and Sam’s clearly superior politicking, led to the success of his subsequent coup against the board, backed by Microsoft. This created another ripple effect that ultimately led to the exodus of additional remaining high-profile employees, now including Ilya Sutskever and Mira Murati, among a long list of others. Disillusioned by Altman’s direction, many of these former executives and engineers have gone on to form or join competitors.

There are obviously different ways to tell the story here, and I don’t think it’s clear which is more true, or which provides more insight - but both seem worth telling.

The first is that a breakthrough technology like LLMs and the success of scaling was never going to stay within a single nonprofit. This telling is about economic forces; the commercial implications were too lucrative for any mission focused on the social good to eventually win-out over the immense sums of money that are now at stake. A similar claim has been made about the national security implications of the technology, and the inevitability of national government takeover of AI. The latter is still to be seen, but is likely to depend on how quickly or slowly the technology advances, and how fast others are to see what is happening. And counterfactually we can wonder the same about what OpenAI would have looked like if they had stayed focused on safe AI development, rather than AI deployment.

But another way to tell the story is a much older tale, one about hubris and the inevitable end of those who betray their friends. As Altman Said, OpenAI is nothing without its people” - but of the 10 other co-founders, only Greg Brockman and Wojciech Zaremba remain. Altman’s inability to maintain trust with his team, his unwillingness to stay true to OpenAI’s original mission, and his failure to foster an environment where the worlds best AI researchers wanted to stay, all opened the door for rival organizations to capitalize on AI advancements, and potentially squandered the single best chance humanity had of investing in AI safety before the current AI capabilities races took over. What was once a monumental head start for those who wanted safety and prosperity over profit is gone. And the creation of competitors—powered largely by former OpenAI insiders—has led to what is likely the single biggest fumbled technology lead in history, if not also a far larger failure.