He built Cohere into a $5.5B AI startup; How to Win in AI; & Why LLMs won't lead to AGI. | Nick Frosst, Co-Founder of Cohere
Sep 9, 2024
auto_awesome
Nick Frosst, co-founder of the $5.5B AI startup Cohere, shares insights from his journey in the enterprise AI landscape. He argues that while large language models (LLMs) are powerful, they won't achieve artificial general intelligence. Nick emphasizes the importance of solving real problems rather than leveraging AI for its own sake and considers ChatGPT a UI/UX revolution. He offers advice for founders about navigating intense competition and the need for a niche to thrive in the AI ecosystem.
Nick Frosst emphasizes that while large language models are powerful, they have significant limitations and will not lead to artificial general intelligence.
Founders are advised to manage emotional stress effectively, recognizing that personal resilience is crucial in navigating the entrepreneurial journey.
Cohere distinguishes itself in the AI market by focusing on enterprise solutions that address specific business needs rather than pursuing broader consumer applications.
Deep dives
Skepticism Towards AGI Development
There is a strong belief that developing Artificial General Intelligence (AGI) is not an immediate goal and might not even be desired. The speaker expresses that while the advancements in AI technology are exciting, there is skepticism about the need for computers to emulate human-like intelligence. They emphasize that current models are primarily sequence models that predict words based on training data. This indicates a recognition of the limitations of current AI technology and a focus on practical applications rather than pursuing the elusive goal of AGI.
Emotional Challenges for Founders
Founders often experience intense emotional stress as they navigate their entrepreneurial journey, regardless of the company's size. The podcast highlights that each new challenge, no matter how seemingly small, feels significant at the moment, and that perspective tends to change over time. Founders are encouraged to remain calm and manage their stress effectively, understanding that the emotional challenges will persist as the business grows. This advice serves as a reminder that the journey of building a startup is as much about personal resilience as it is about business development.
Surprises in AI's Data Efficiency
One of the most surprising insights shared is the impact of reinforcement learning from human feedback on AI models. The speaker explains that it was unexpected how a small amount of human feedback could drastically improve the performance and usability of large language models. This revelation changed the expectations regarding how data efficiency could enhance model interactions. The ability to fine-tune models effectively with less data has shifted the paradigm of what is feasible in AI development.
The Evolution of AI Usability
The discussion brings attention to the significant increase in usability of AI models since the introduction of user-friendly interfaces, particularly with the advent of chat models like ChatGPT. Previously, users needed to carefully engineer prompts to elicit useful responses from AI, which limited accessibility. With the new advancements, AI has become much easier to work with, enabling a broader audience to benefit from its capabilities without requiring extensive technical knowledge. This evolution reflects a shift toward more intuitive interaction with AI technology, making it applicable across various industries.
Focusing on Enterprise Applications
The podcast emphasizes the strategic direction of focusing on enterprise solutions rather than consumer applications, which differentiates certain AI companies in a crowded market. The speaker highlights that businesses are prioritizing the practical utility of language models for internal tasks and external customer interactions. Companies need models that can securely handle proprietary data, understand multi-language contexts, and generate useful outputs for decision-making. This pragmatic approach focuses on creating functional tools that address specific business needs, rather than pursuing broader but less applicable consumer experiences.
In this episode, I sit down with Nick Frosst, Co-Founder of Cohere, the $5.5B AI startup that’s targeting the enterprise landscape.
We go through the origin story of Cohere, the challenges of building foundational models, and why he believes large language models (LLMs) won’t lead to artificial general intelligence (AGI). We also explore the fierce competition in AI, what sets Cohere apart, and Nick’s advice for founders building in AI today.
Why you should listen
LLMs are powerful but have clear limitations and won't lead to AGI.
Why AI startups need to start with real problems vs leveraging AI for its own sake
Why ChatGPT was as much of a UI/UX revolution than a technological one
What tech founders need to do to win in AI
Timestamps:
(00:00:00) Intro (00:03:21) AI Expectations (00:06:05) A Unique and New Moment (00:09:38) Resource Intensive Industry (00:12:03) Zero to One (00:15:07) Base Language Model to Chat Model (00:17:15) Carving Out a Niche (00:21:03) Open Source (00:24:00) The Limits of LLMs (00:26:18) Agents (00:29:30) AGI (00:34:04) A Little Bit of Data (00:39:05) Speed of Development (00:40:47) Finding True Product Market Fit (00:43:37) One Piece of Advice