YouTube Link: https://www.youtube.com/watch?v=41JBrC5e5tA

David Hand, professor of statistics, reveals how ChatGPT lies with "dark data"; more generally, large language models and even peer review.

Listen now early and ad-free on Patreon https://patreon.com/curtjaimungal.

- Patreon: https://patreon.com/curtjaimungal (early access to ad-free audio episodes!)

DAVID HAND'S BOOKS:

- The Improbability Principle: https://amzn.to/3DOn1iX

TIMESTAMPS:

00:00:00 Introduction

00:01:34 What is Dark Data? (missing data matters more than what you have)

00:07:03 The perils of "changing definitions"

00:09:15 David on writing and his selective process

00:20:15 Theory-driven vs. data-driven models (& the constitution of LLMs)

00:32:08 The dilemma of partial truths

00:34:40 The "File Drawer Problem" & its adverse effects on clinical trials

00:39:09 Regression to the mean (how random variations lead to misleading conclusions)

00:44:12 Publication bias

00:48:03 Open-access models and their pitfalls

00:54:06 Why LLMs are simultaneously brilliant & stupid

01:03:40 David’s daily routine

01:06:24 The mean vs. median

01:11:07 Every type of "Dark Data" listed (watch this first!)

The Improbability Principle

Mentioned by