All Episodes
Linear Digressions — 327 episodes · page 1 of 4
Back in Two Weeks
Mechanistic Interpretability: How Researchers Try To See Inside Models
LLM As A Judge
The Impact of AI on Podcasting (Harvard Data Science Review Cross-Post)
Better Know A Benchmark: ExploitGym
Constitutional AI
A Data-Driven Reality Check on AI in Business (Interview with Tom Davenport, Babson College)
Understanding AI Text Watermarking
Better Know a Benchmark: Humanity's Last Exam
A Scientific Deep Dive into Overconfident LLMs: Interview with Kaitlyn Zhou (Cornell)
Reasoning Models: When LLMs Went Beyond Fancy Autocomplete
Distillation, or, How to Steal a Model
Invisible LLM Failures and AI Fluency with Chris Potts (Stanford)
Still summer break: back next week
Summer break: back soon
Interviewing the Linear Digressions Agents (The Agents Season, Episode 11)
Agent Economics (The Agents Season, Episode 10)
Agent Trust, Oversight and Control (The Agents Season, Episode 9)
Many Agents, Many Problems (The Agents Season, Episode 8)
How Do You Evaluate An AI Agent? (The Agents Season, Episode 7)
AI Agent Failure Modes (The Agents Season, Episode 6)
Agentic Planning (The Agents Season, Episode 5)
Memory Management for AI Agents (The Agents Season, Episode 4)
Lost in the Middle (The Agents Season, Episode 3)
ReAct and Tool Usage (The Agents Season, Episode 2)
What's an AI Agent? And Why's That Hard to Define? (The Agents Season, Episode 1)
Unfaithful Chain of Thought
Benchmark Bank Heist
Benchmarking AI Models
The Hot Mess of AI (Mis-)Alignment
The Bitter Lesson
From Atari to ChatGPT: How AI Learned to Follow Instructions
It's RAG time: Retrieval-Augmented Generation
Chasing Away Repetitive LLM Responses with Verbalized Sampling
We're Back
A Key Concept in AI Alignment: Deep Reinforcement Learning from Human Preferences
The Impact of Generative AI on Critical Thinking
So long, and thanks for all the fish
A Reality Check on AI-Driven Medical Assistants
A Data Science Take on Open Policing Data
Procella: YouTube's super-system for analytics data storage
The Data Science Open Source Ecosystem
Rock the ROC Curve
Criminology and Data Science
Racism, the criminal justice system, and data science
An interstitial word from Ben
Convolutional Neural Networks
Stein's Paradox
Protecting Individual-Level Census Data with Differential Privacy
Causal Trees
The Grammar Of Graphics
Gaussian Processes
Keeping ourselves honest when we work with observational healthcare data
Changing our formulation of AI to avoid runaway risks: Interview with Prof. Stuart Russell
Putting machine learning into a database
The work-from-home episode
Understanding Covid-19 transmission: what the data suggests about how the disease spreads
Network effects re-release: when the power of a public health measure lies in widespread adoption
Causal inference when you can't experiment: difference-in-differences and synthetic controls
Better know a distribution: the Poisson distribution
The Lottery Ticket Hypothesis
Interesting technical issues prompted by GDPR and data privacy concerns
Thinking of data science initiatives as innovation initiatives
Building a curriculum for educating data scientists: Interview with Prof. Xiao-Li Meng
Running experiments when there are network effects
Zeroing in on what makes adversarial examples possible
Unsupervised Dimensionality Reduction: UMAP vs t-SNE
Data scientists: beware of simple metrics
Communicating data science, from academia to industry
Optimizing for the short-term vs. the long-term
Interview with Prof. Andrew Lo, on using data science to inform complex business decisions
Using machine learning to predict drug approvals
Facial recognition, society, and the law
Lessons learned from doing data science, at scale, in industry
Varsity A/B Testing
The Care and Feeding of Data Scientists: Growing Careers
The Care and Feeding of Data Scientists: Recruiting and Hiring Data Scientists
The Care and Feeding of Data Scientists: Becoming a Data Science Manager
Procella: YouTube's super-system for analytics data storage
Kalman Runners
What's *really* so hard about feature engineering?
Data storage for analytics: stars and snowflakes
Data storage: transactions vs. analytics
GROVER: an algorithm for making, and detecting, fake news
Data science teams as innovation initiatives
Can Fancy Running Shoes Cause You To Run Faster?
Organizational Models for Data Scientists
Data Shapley
A Technical Deep Dive on Stanley, the First Self-Driving Car
An Introduction to Stanley, the First Self-Driving Car
Putting the "science" in data science: the scientific method, the null hypothesis, and p-hacking
Interleaving
Federated Learning
Endogenous Variables and Measuring Protest Effectiveness
Deepfakes
Revisiting Biased Word Embeddings
Attention in Neural Nets
Interview with Joel Grus