Sara Hooker, the former Vice President of AI Research at Cohere, is launching a really cool new venture. Her ultimate aspiration is to change the way artificial intelligence grows, evolves and adapts to real-world experiences. With Sudip Roy, another fellow veteran from Cohere and Google, she co-founded Adaption Labs. Logan AI’s mission is to develop AI systems that always learn and adapt in the most efficient way possible.
Through its Adaption Labs, Adaption specializes in closing a vital gap in today’s AI models. Yet, according to Hooker, these models frequently fail to train robustly on lived experiences from the world. This gap in capability presents a major challenge for safety and generalization of AI technologies across complex environments.
Hooker’s AI research experience extends well past her stint at Cohere. During her time as an alumna of Google Brain, she learned monumental lessons about creating cutting-edge AI systems. It’s clear that she has spent her life working to broaden access to cutting-edge AI research globally. Her work today now includes aggressively recruiting talent from underrepresented continents, such as Africa.
Their fundworks model is currently looking to raise between $20 million and $40 million. They have begun to actively raise a seed funding round for this fall. This funding will supercharge their mission to responsibly develop AI systems that continuously adapt and learn from real-world interactions. Instead they’re beginning to abandon the one-size-fits-all scaling approach that’s ruled the industry for decades.
Hooker is worried about the direction of thinking that has taken hold—that simply making models bigger will, by itself, produce more intelligent systems. She stated, “There is a turning point now where it’s very clear that the formula of just scaling these models — scaling-pilled approaches, which are attractive but extremely boring — hasn’t produced intelligence that is able to navigate or interact with the world.”
Her ideas resonate strongly with Richard Sutton’s, another Turing Award winner. For one, he’s taken issue with the scalability of so-called large language models (LLMs) because they cannot learn from real-world experience. Despite billions being invested in scaling LLMs under the assumption that size equates to general intelligence, both Hooker and Sutton advocate for a shift towards models that can dynamically learn and adapt in context.
Hooker further elaborated on her vision for Adaption Labs, stating, “We have a handful of frontier labs that determine this set of AI models that are served the same way to everyone, and they’re very expensive to adapt. I think that doesn’t need to be true anymore, and AI systems can very efficiently learn from an environment. Proving that will completely change the dynamics of who gets to control and shape AI, and really, who these models serve at the end of the day.”
This one-sided traditional approach to AI development has heavily benefited a narrow set of organizations. These philanthropic giants rule the play field with their gigantic grants and talents. Hooker’s initiative aims to democratize access to, and control of, AI technologies, with the potential to reshape competitive dynamics within the industry.
Adaption Labs is a major shift away from traditional processes by focusing on getting the most value possible out of learning from the real world. This creative shift can provide new pathways for institutions that have been deeply rooted in the paradigm of large dollar commitments to perfecting previous generations. For instance, OpenAI reportedly requires clients to spend upwards of $10 million to receive consulting services focused on model adaptation.
As exciting as this new development is, some industry leaders remain unconvinced that RL could significantly advance AI models in the long run. It is Hooker’s emphasis on adaptive learning that may result in the creation of more robust and ubiquitous AI agents.
