Scaling enterprise AI: Fireside chat with Eli Lilly’s Diogo Rau and Dario Amodei

Key Concepts

• Enterprise AI Strategy: How AI providers tailor their offerings for businesses, contrasting with consumer AI.
• Model Sycophancy: A tendency for AI models to agree with users, even when incorrect, a problem for enterprise applications.
• Accuracy and Reliability: Crucial metrics for enterprise AI, especially in high-stakes fields like drug discovery.
• Specialized Claudes: AI models fine-tuned or augmented with specific knowledge bases for particular industries (e.g., Financial Services, Life Sciences).
• End-to-End AI Deployment: The goal of AI models capable of handling entire processes without human intervention.
• Parallel Development: The strategy of preparing for future AI capabilities while current models are still improving, to accelerate deployment.

Anthropic's Enterprise Strategy and Differentiation

Dario, Founder and CEO of Anthropic, outlines Anthropic's distinct approach to enterprise AI, contrasting it with consumer AI. The primary driver for consumer AI is engagement and growth, which can lead to undesirable behaviors like "model sycophancy." This is where models tend to agree with users, even if the information is incorrect. While this can be problematic for consumers, it poses significantly greater risks for enterprises.

Key Points:

• Contrast with Consumer AI: Consumer AI incentives focus on engagement and growth, leading to behaviors like sycophancy.
• Enterprise Need for Truth: In enterprise settings, particularly in fields like drug discovery, accuracy and reliability are paramount. Enterprises require models that provide truthful and dependable information, not just agreeable responses.
• Anthropic's Design Philosophy: Anthropic's model design prioritizes making models smarter and better at economically valuable tasks, with a strong emphasis on accuracy and reliability.
• Value of Specialized Knowledge: Dario highlights that improvements in specialized knowledge, such as biochemistry at a graduate level, are highly valued by enterprises, unlike the general consumer who might not perceive the difference.

Specialized Claudes and Industry-Specific Applications

Anthropic is developing specialized versions of their Claude models, tailored for specific industries. These can involve either inherent improvements to the model's capabilities or augmenting the model with access to relevant information.

Examples and Applications:

• Claude for Financial Services: This version is connected to financial indices and ratings, making it aware of and able to utilize this specialized knowledge.
• Claude for Life Sciences: Anthropic is working on a version for life sciences that will combine inherent model improvements with access to vast databases of biological information, such as proteins, compounds, and assays. This aims to provide the model with critical data at its "fingertips."

Advice for Drug Discovery and Development Professionals

Dario offers crucial advice for those working in drug discovery and development, emphasizing ambition and foresight regarding AI's future capabilities.

Key Arguments and Perspectives:

• Avoid Incrementalism: There's a temptation to start with small AI applications within existing processes. However, this can be challenging as it requires integrating AI into specific parts of a complex, multi-step workflow, where non-AI parts might create dependencies.
• Embrace End-to-End AI: Professionals should be ambitious and prepare for AI models that can handle entire processes end-to-end.
• Faith in Technological Progress: Dario urges faith in the rapid pace of AI development. If models become capable of end-to-end execution in a year, delaying deployment until then will result in a further two-year delay, during which patient benefits are lost.
• Parallel Development Strategy: The recommended approach is to start preparing for large-scale AI integration in parallel with model development. This proactive stance can save years of time.
• Courage and Foresight: Implementing this parallel strategy requires courage and foresight, as it involves planning for a future state of AI capabilities rather than relying on current limitations.

Notable Statements:

• "Let's have faith in the pace of progress of the technology, because if the models get good enough to do it end to end, a year from now and only then you start deploying it. There will be another two year delay and that's, you know, that's two years during which all the work that you're doing to benefit patients is not happening." - Dario
• "Whereas if you go in parallel, if you start preparing now for the large change as the models are getting better, then you know, you may save years of time." - Dario
• "So don't do two year long projects and expect that it's gonna be exactly the same way in two years from now as this." - Diogo Rau
• "Yes, if you do two year long projects, plan for where the AI is gonna, I mean, that sounds like an obvious thing to say, but I think it actually takes a lot of courage and foresight to do that." - Diogo Rau

Conclusion and Main Takeaways

The discussion between Diogo Rau of Eli Lilly and Dario of Anthropic highlights a critical shift in AI strategy for enterprises. Anthropic differentiates itself by prioritizing accuracy and reliability over engagement, making its models more suitable for high-stakes industries like pharmaceuticals. The development of specialized Claudes, such as for Life Sciences, signifies a move towards deeply integrated AI solutions. The core takeaway for professionals in drug discovery and development is to be ambitious, embrace the potential for end-to-end AI capabilities, and adopt a parallel development strategy. This proactive approach, rather than waiting for AI to mature before deployment, is essential to accelerate innovation and deliver benefits to patients sooner. The conversation underscores the need for courage and foresight in planning for the rapid evolution of AI technology.