Ep 77: Anthropic’s Dianne Na Penn on Opus 4.5, Rethinking Model Scaffolding & Safety as a Competitive Advantage

In this episode of Unsupervised Learning, Jacob Effron interviews Dianne Na Penn, Head of Product for Research at Anthropic, about the new Opus 4.5 model. They discuss the model's improved capabilities, including instruction following, coding, and memory, as well as its applications in areas like Excel and PowerPoint. Dianne shares insights into Anthropic's research process, the importance of safety in product development, and the company's focus on business use cases. They explore the potential of computer use and long-running AI agents, and the need for better evaluations to measure the quality of AI judgment. Dianne also touches on Anthropic's culture, key decision points in their journey, and the benefits of aligned AI models.

Outlines

Sign in to continue reading, translating and more.

Continue

Unsupervised Learning with Jacob Effron

Introduction to Opus 4.5 and Anthropic's Model Development Process

Envisioning Future Capabilities and Computer Use Advancements in Opus 4.5

Internal Use and Efficiency Gains of Opus 4.5

Educating the Market and Early Surprises with Opus 4.5

Product Market Fit and Prioritization of Model Improvements

Feedback Loops, Internal Tooling, and the Discourse on Enterprise Agents

Anthropic's Perspective on AI Transformation and Future Directions

Long-Running Agents and the Need for Better Evaluations

Evolving Evaluations and the Importance of End-User Measurements

The Evolution of Scaffolding and Harnesses Around Models

Anthropic's Culture and Dianne's Day-to-Day Role

Key Decision Points and Fun Moments at Anthropic

Quick Fire Round: Changing Minds, Sophisticated Users, and Model Taste

Hackathons, Under-Discussed Implications, and ASI Timelines

Ep 77: Anthropic’s Dianne Na Penn on Opus 4.5, Rethinking Model Scaffolding & Safety as a Competitive Advantage

Unsupervised Learning with Jacob Effron

00:00Introduction to Opus 4.5 and Anthropic's Model Development Process

Introduction to Opus 4.5 and Anthropic's Model Development Process

04:22Envisioning Future Capabilities and Computer Use Advancements in Opus 4.5

Envisioning Future Capabilities and Computer Use Advancements in Opus 4.5

07:15Internal Use and Efficiency Gains of Opus 4.5

Internal Use and Efficiency Gains of Opus 4.5

09:12Educating the Market and Early Surprises with Opus 4.5

Educating the Market and Early Surprises with Opus 4.5

12:54Product Market Fit and Prioritization of Model Improvements

Product Market Fit and Prioritization of Model Improvements

14:40Feedback Loops, Internal Tooling, and the Discourse on Enterprise Agents

Feedback Loops, Internal Tooling, and the Discourse on Enterprise Agents

17:18Anthropic's Perspective on AI Transformation and Future Directions

Anthropic's Perspective on AI Transformation and Future Directions

20:09Long-Running Agents and the Need for Better Evaluations

Long-Running Agents and the Need for Better Evaluations

23:11Evolving Evaluations and the Importance of End-User Measurements

Evolving Evaluations and the Importance of End-User Measurements

25:05The Evolution of Scaffolding and Harnesses Around Models

The Evolution of Scaffolding and Harnesses Around Models

27:59Anthropic's Culture and Dianne's Day-to-Day Role

Anthropic's Culture and Dianne's Day-to-Day Role

30:28Key Decision Points and Fun Moments at Anthropic

Key Decision Points and Fun Moments at Anthropic

33:45Quick Fire Round: Changing Minds, Sophisticated Users, and Model Taste

Quick Fire Round: Changing Minds, Sophisticated Users, and Model Taste

37:17Hackathons, Under-Discussed Implications, and ASI Timelines

Hackathons, Under-Discussed Implications, and ASI Timelines