Mira Murati, OpenAI’s Chief Technology Officer, announced her departure along with two other executives, marking a significant leadership turnover as the company seeks new funding amid an AI arms race. Murati, who joined in 2018, cited a desire for personal exploration as her reason for leaving. Her exit follows a series of departures among senior leaders. It comes at a critical time as OpenAI plans to restructure from a nonprofit to a for-profit benefit corporation while pursuing over $100 billion in investment.
I’ll highlight a bit of analysis in Blood in the Machine. Quote “Eventually the chickens come home to roost, and the industry will likely be forced to admit that it doesn’t make sense to have a dozen different firms running very similar and highly intensive models to produce text and image output. OpenAI enjoys name recognition, that prime mover status. And venture capital and well-heeled backers kept Uber afloat for over a decade before it got anywhere near turning a profit. But generative AI is a much more resource-intensive and expensive technology—how long will those investors and oddly structured partnerships tolerate such high costs, in the event that gen AI’s use cases wind up much more limited in a business context than currently promised? Generative AI is not going to disappear like NFTs, but we can expect a massive ‘right-sizing’ of the industry in coming years to be sure.”
Hugging Face has surpassed 1 million AI model listings, reflecting rapid growth in machine learning. The platform, which evolved from a chatbot app to an AI model hub, emphasizes the importance of customized models for specific tasks. The increase in models is driven by fine-tuning existing models and collaborative contributions from developers worldwide.
So, let’s throw in a detail from last week’s Meta Connect event. Mark Zuckerberg stated that many creators overestimate the value of their work for AI training, suggesting that Meta will only pay for content deemed valuable. He highlighted ongoing copyright issues related to data scraping for AI. He indicated that Meta may form partnerships for important content but would refrain from using creators’ work if they object.
Let’s combine two insights here. First, Hugging Face indicates that a lot of those models are private, one user ones, meaning they are customized for a particular organization or user. At first glace you might think there are that many models – no, there aren’t. The push for specialized, fine-tuned models directly responds to the inefficiency of one-size-fits-all solutions as the industry moves toward a more modular approach.
Now we combine that with the instability of OpenAI’s leadership – indicative of a startup having growing pains, and the potential right sizing of this market, where we could see consolidation, market exits, or a reduction in the number of players trying to sell AI solutions. The industry will likely shift towards practical AI applications with demonstrable ROI, rather than general-purpose solutions. That’s a good place for service providers to be.
I also threw in Zuckerberg’s comments to remind us that we may be overvaluing the datasets. It’s only worth what someone will pay for it.