#316 Robbie Goldfarb: Why the Future of AI Depends on Better Judgment

The podcast explores the critical role of expert evaluation in improving AI systems, particularly in sensitive areas like healthcare and politics. Robbie Goldfarb, founder of Forum AI, discusses their approach to scaling expert knowledge to evaluate AI, emphasizing the need for credible experts to shape AI development. Forum AI captures expert thought processes through detailed interviews and consequence mapping, building AI judges that mirror expert reasoning. The discussion highlights the challenge of aligning AI with subjective domains where ground truth is elusive, advocating for transparency and diverse expert networks. Goldfarb addresses the importance of trust in AI, especially regarding mental health and political discourse, and the necessity of managing user expectations around AI's capabilities.

Outlines

Part 1: Introduction, Context

Part 2: Methodology, Expert Knowledge

Part 3: Standards, Bias, Reliability

Part 4: Trust, Responsibility

Part 5: Industry Challenges, Business Model

Part 6: Comparisons, Conclusion

Sign in to continue reading, translating and more.

Continue

Eye On A.I.

Part 1: Introduction, Context

AI's Impact on Society: Balancing Trust, Misinformation, and Shaping Outcomes

Forum AI: Scaling Expert Evaluation to Improve AI Systems in Contentious Spaces

Part 2: Methodology, Expert Knowledge

Extracting and Generalizing Expert Knowledge for AI Judges: The Thought Process Approach

Navigating AI Truth: Balancing Subjectivity, User Intent, and Evaluation Dimensions

Capturing Expert Thought Processes: Interviews, Consequence Mapping, and Data Labeling

Part 3: Standards, Bias, Reliability

Defining Trusted Sources: Balancing Credibility, Transparency, and Differing Opinions in AI

Transparency and Edge Cases: Protecting Against Bias and Fine-Tuning AI Models

Evaluating AI: Bias, Factuality, Source Selection, and Tone in Political Topics

Expert Network: Quality over Quantity in AI Evaluation and the Ground Truth

Extracting Expert Knowledge from Writings: Themes, Judgments, and Training Judges

Part 4: Trust, Responsibility

Building Trust in AI: Bridging the Gap Between Optimism and Reliability

Trust in AI: Verifiable Scenarios, Unverifiable Scenarios, and Responsibility

Addressing Sycophancy and Source Selection: Improving AI Responses on Contentious Topics

Defining Good Judgment in AI: Handling Unverifiable Scenarios and Expert Guidance

Part 5: Industry Challenges, Business Model

Critical Issues Facing AI: Mental Health, Politics, and the Health of Information

Forum AI's Customers: AI Labs, System Integrators, and the Business Model

Sovereign AI Models: Objectivity, Cultural Views, and Transparency

Part 6: Comparisons, Conclusion

Delphi vs. Forum AI: Cloning Experts for Evaluation vs. Conversation

Public Evaluations and Leaderboards: Balancing Transparency and Integrity

Nuance in AI Models: No Model is Universally Better, Contact Information

#316 Robbie Goldfarb: Why the Future of AI Depends on Better Judgment

Eye On A.I.

Part 1: Introduction, Context

00:00AI's Impact on Society: Balancing Trust, Misinformation, and Shaping Outcomes

AI's Impact on Society: Balancing Trust, Misinformation, and Shaping Outcomes

01:49Forum AI: Scaling Expert Evaluation to Improve AI Systems in Contentious Spaces

Forum AI: Scaling Expert Evaluation to Improve AI Systems in Contentious Spaces

Part 2: Methodology, Expert Knowledge

05:02Extracting and Generalizing Expert Knowledge for AI Judges: The Thought Process Approach

Extracting and Generalizing Expert Knowledge for AI Judges: The Thought Process Approach

09:40Navigating AI Truth: Balancing Subjectivity, User Intent, and Evaluation Dimensions

Navigating AI Truth: Balancing Subjectivity, User Intent, and Evaluation Dimensions

13:41Capturing Expert Thought Processes: Interviews, Consequence Mapping, and Data Labeling

Capturing Expert Thought Processes: Interviews, Consequence Mapping, and Data Labeling

Part 3: Standards, Bias, Reliability

17:10Defining Trusted Sources: Balancing Credibility, Transparency, and Differing Opinions in AI

Defining Trusted Sources: Balancing Credibility, Transparency, and Differing Opinions in AI

21:11Transparency and Edge Cases: Protecting Against Bias and Fine-Tuning AI Models

Transparency and Edge Cases: Protecting Against Bias and Fine-Tuning AI Models

25:32Evaluating AI: Bias, Factuality, Source Selection, and Tone in Political Topics

Evaluating AI: Bias, Factuality, Source Selection, and Tone in Political Topics

28:13Expert Network: Quality over Quantity in AI Evaluation and the Ground Truth

Expert Network: Quality over Quantity in AI Evaluation and the Ground Truth

32:15Extracting Expert Knowledge from Writings: Themes, Judgments, and Training Judges

Extracting Expert Knowledge from Writings: Themes, Judgments, and Training Judges

Part 4: Trust, Responsibility

35:58Building Trust in AI: Bridging the Gap Between Optimism and Reliability

Building Trust in AI: Bridging the Gap Between Optimism and Reliability

38:27Trust in AI: Verifiable Scenarios, Unverifiable Scenarios, and Responsibility

Trust in AI: Verifiable Scenarios, Unverifiable Scenarios, and Responsibility

40:22Addressing Sycophancy and Source Selection: Improving AI Responses on Contentious Topics

Addressing Sycophancy and Source Selection: Improving AI Responses on Contentious Topics

45:01Defining Good Judgment in AI: Handling Unverifiable Scenarios and Expert Guidance

Defining Good Judgment in AI: Handling Unverifiable Scenarios and Expert Guidance

Part 5: Industry Challenges, Business Model

46:11Critical Issues Facing AI: Mental Health, Politics, and the Health of Information

Critical Issues Facing AI: Mental Health, Politics, and the Health of Information

49:58Forum AI's Customers: AI Labs, System Integrators, and the Business Model

Forum AI's Customers: AI Labs, System Integrators, and the Business Model

52:58Sovereign AI Models: Objectivity, Cultural Views, and Transparency

Sovereign AI Models: Objectivity, Cultural Views, and Transparency

Part 6: Comparisons, Conclusion

55:50Delphi vs. Forum AI: Cloning Experts for Evaluation vs. Conversation

Delphi vs. Forum AI: Cloning Experts for Evaluation vs. Conversation

57:52Public Evaluations and Leaderboards: Balancing Transparency and Integrity

Public Evaluations and Leaderboards: Balancing Transparency and Integrity

1:01:07Nuance in AI Models: No Model is Universally Better, Contact Information

Nuance in AI Models: No Model is Universally Better, Contact Information