Former Cohere AI Research Lead Launches Adaption Labs to Challenge Scaling Paradigm
Key Points
- Sara Hooker leaves Cohere to launch AI startup Adaption Labs with Sudip Roy.
- Adaption Labs focuses on building AI that continuously adapts and learns from real‑world experience.
- The company critiques the industry’s reliance on ever‑larger language models, citing diminishing returns.
- Seed funding has been secured, though the exact amount has not been disclosed.
- Adaption Labs plans to open a San Francisco office while hiring talent globally.
- The venture reflects a broader shift among researchers toward adaptive learning and efficiency.
- If successful, the approach could lower costs and expand access to advanced AI capabilities.
Sara Hooker, a former vice president of AI research at Cohere and former Google Brain researcher, has quietly launched a new startup called Adaption Labs with fellow AI veteran Sudip Roy. The company aims to build artificial‑intelligence systems that continuously adapt and learn from real‑world experience, arguing that the industry’s focus on ever‑larger language models is reaching diminishing returns. Hooker’s critique of the "scaling" approach echoes a growing chorus of researchers who see adaptive learning as a more efficient path forward. Adaption Labs has secured seed funding and plans to hire globally while opening a San Francisco office.
Background and Motivation
In recent years, leading AI laboratories have poured billions of dollars into building data centers the size of Manhattan, driven by a belief that scaling up compute and model size will eventually produce superintelligent systems. However, a number of researchers have begun to question whether simply enlarging large language models (LLMs) can continue to deliver meaningful gains. Sara Hooker, who previously led Cohere’s AI research division and spent time at Google Brain, has been vocal about the limits of the scaling‑only strategy.
Hooker argues that the “formula of just scaling these models” has not produced intelligence capable of navigating real‑world environments. She points to the need for AI systems that can learn from experience in the same way humans do—adjusting behavior after a mistake, such as stepping more carefully around a familiar obstacle.
Founding of Adaption Labs
In response to these concerns, Hooker and Sudip Roy, another veteran of Cohere and Google, co‑founded Adaption Labs. The startup’s mission is to develop AI that can continuously adapt and learn from real‑world interactions, doing so efficiently without relying solely on massive pre‑training datasets. While the company has not disclosed the specific technical approaches it will use, Hooker emphasizes that adaptive learning could dramatically reduce the cost and energy consumption associated with today’s scaling‑focused models.
Adaption Labs announced its formation quietly, using a brief social‑media post to signal its hiring plans. The company intends to open an office in San Francisco while recruiting talent worldwide, reflecting Hooker’s previous commitment to building diverse research teams across under‑represented regions.
Industry Context and Reactions
The launch of Adaption Labs arrives amid growing skepticism within the AI community. Recent academic papers have suggested that the performance gains from ever‑larger models are diminishing. Prominent researchers, including a Turing Award winner known for reinforcement learning, have publicly questioned whether LLMs can truly scale without incorporating real‑world learning capabilities.
Companies such as OpenAI have begun exploring alternative approaches, like AI reasoning models that take additional time to solve problems before answering. Nonetheless, the prevailing industry narrative still heavily emphasizes scaling compute and data. Hooker’s venture represents a direct challenge to that narrative, proposing that adaptive learning could democratize AI development by lowering barriers to entry.
Funding and Future Outlook
Adaption Labs has reportedly closed a seed funding round, though the exact amount remains undisclosed. Hooker indicated that the company is “set up to be very ambitious,” signaling confidence in its ability to attract further investment. The startup plans to focus on building compact AI systems that can outperform larger counterparts on tasks such as coding, mathematics, and reasoning.
If Hooker’s vision proves successful, it could reshape the AI landscape by shifting emphasis away from sheer model size toward efficiency and adaptability. The implications extend beyond technology, potentially influencing how AI is governed, who has access to advanced models, and how future AI systems interact with the real world.