Published Dec 23, 2022

#92 - SARA HOOKER - Fairness, Interpretability, Language Models

Sara Hooker delves into the frontier of AI ethics, highlighting challenges in ensuring fairness and addressing bias in machine learning models. The discussion encapsulates the complexity of language models, emphasizing interpretability, adaptive computation, and the importance of aligning AI systems with human values.

Episode Highlights

Topics covered

Episode Highlights

Related Episodes

Sara Hooker - The Hardware Lottery, Sparsity and Fairness
Answers 383 questions
Sara Hooker - Why US AI Act Compute Thresholds Are Misguided
Answers 383 questions
#73 - YASAMAN RAZEGHI & Prof. SAMEER SINGH - NLP benchmarks
Answers 383 questions
#039 - Lena Voita - NLP
Answers 383 questions
047 Interpretable Machine Learning - Christoph Molnar
Answers 383 questions
#114 - Secrets of Deep Reinforcement Learning (Minqi Jiang)
Answers 383 questions
#032- Simon Kornblith / GoogleAI - SimCLR and Paper Haul!
Answers 383 questions
Explainability, Reasoning, Priors and GPT-3
Answers 383 questions
#046 The Great ML Stagnation (Mark Saroufim and Dr. Mathew Salvaris)
Answers 383 questions
#55 Self-Supervised Vision Models (Dr. Ishan Misra - FAIR).
Answers 383 questions
#48 Machine Learning Security - Andy Smith
Answers 383 questions
#80 AIDAN GOMEZ [CEO Cohere] - Language as Software
Answers 383 questions
#68 DR. WALID SABA 2.0 - Natural Language Understanding [UNPLUGGED]
Answers 383 questions
#91 - HATTIE ZHOU - Teaching Algorithmic Reasoning via In-context Learning #NeurIPS
Answers 383 questions
#040 - Adversarial Examples (Dr. Nicholas Carlini, Dr. Wieland Brendel, Florian Tramèr)
Answers 383 questions

Dexa/Machine Learning Street Talk (MLST)

#92 - SARA HOOKER - Fairness, Interpretability, Language Models

Topics covered

Popular Clips

Model Interpretability Insights

Adaptive Learning Challenges

Error Distinction Insights

Navigating Prompt Engineering

Language Model Insights

RLHF Insights

Data vs. Model

Tyranny of Objectives

Understanding Toxicity Labels

Model Optimization Insights

Evolving Fairness in AI

Prompting Dynamics

Simplifying Neural Inputs

Adaptive Learning Strategies

Episode Highlights

Fairness and Bias

Language Model Dynamics

Interpretability and Optimization

Related Episodes

Sara Hooker - The Hardware Lottery, Sparsity and Fairness

Sara Hooker - Why US AI Act Compute Thresholds Are Misguided

#73 - YASAMAN RAZEGHI & Prof. SAMEER SINGH - NLP benchmarks

#039 - Lena Voita - NLP

047 Interpretable Machine Learning - Christoph Molnar

#114 - Secrets of Deep Reinforcement Learning (Minqi Jiang)

#032- Simon Kornblith / GoogleAI - SimCLR and Paper Haul!

Explainability, Reasoning, Priors and GPT-3

#046 The Great ML Stagnation (Mark Saroufim and Dr. Mathew Salvaris)

#55 Self-Supervised Vision Models (Dr. Ishan Misra - FAIR).

#48 Machine Learning Security - Andy Smith

#80 AIDAN GOMEZ [CEO Cohere] - Language as Software

#68 DR. WALID SABA 2.0 - Natural Language Understanding [UNPLUGGED]

#91 - HATTIE ZHOU - Teaching Algorithmic Reasoning via In-context Learning #NeurIPS

#040 - Adversarial Examples (Dr. Nicholas Carlini, Dr. Wieland Brendel, Florian Tramèr)

#92 - SARA HOOKER - Fairness, Interpretability, Language Models

Topics covered

Popular Clips

Episode Highlights

Fairness and BiasSara Hooker, a leading figure in machine learning, discusses the evolving challenges of fairness in AI and the complexities of model bias. She also explores the distinction between types of errors in AI systems and their implications for data quality.

Fairness and Bias

Language Model DynamicsSara Hooker explores the intricacies of prompting dynamics, adaptiveness challenges, and RLHF in language models. She emphasizes the need for improved understanding and methodologies to enhance model performance and alignment with human values.

Language Model Dynamics

Interpretability and OptimizationSara Hooker explores the intricacies of interpretability techniques and ensemble methods in AI models. She advocates for adaptive computation and efficient data distribution strategies to enhance model performance and fairness.

Interpretability and Optimization

Related Episodes