Scaling Laws and Model Optimization in AI

My guest today is ⁠Sara Hooker⁠, VP of Research at Cohere, where she leads Cohere for AI, a non-profit research lab that seeks to solve complex machine learning problems with researchers from over 100 countries. Sara is the author of numerous research papers, some of which focus specifically on scaling theory in AI. She has been listed as one of AI’s top 13 innovators by Fortune.

In our conversation, we first delve into the scaling laws behind foundation models. We explore what powers the scaling of AI systems and the limits to scaling laws. We then move on to discussing openness in AI, Cohere’s business strategy, the power of ecosystems, the importance of building multilingual LLMs, and the recent change in terms of access to data in the space. I hope you enjoy our conversation.

Find me on X at @⁠⁠ProfSchrepel⁠⁠⁠⁠⁠. Also, be sure to subscribe.

**

References:

➝ Sara Hooker, On the Limitations of Compute Thresholds as a Governance Strategy (2024)

➝ Sara Hooker, The Hardware Lottery (2020)

➝ Sara Hooker, Moving beyond “algorithmic bias is a data problem” (2021)➝ Longpre et al., Consent in Crisis: The Rapid Decline of the AI Data Commons (2024)