The Daniel Stih Podcast

Where did AI companies obtain the actual music recordings used to train their models?

In an earlier investigation, I discovered that some music databases described as containing stolen music actually contained metadata—song titles, artist names, streaming counts—not the recordings themselves.

That left me with a question I couldn't answer: Where did the actual audio files come from?

Recent copyright litigation involving Anthropic and OpenAI revealed how companies obtained millions of books from pirated sources. Could something similar have happened with music?

I explore several possibilities, including employees bringing collections of pirated MP3s to work and the use of tools such as yt-dlp to download audio from YouTube.

I also examine digital rights management (DRM), why copying protected streams can be more difficult, and how Garth Brooks' decision to limit his music's distribution offers an interesting perspective on protecting artists' recordings.

Some AI music copyright lawsuits have settled, while others continue. But important questions remain about where the training recordings came from, how they were acquired, and whether the companies had permission to use them.

If you know how AI companies obtained the actual recordings used to train their music models, I'd like to hear from you.

"Why did people stop liking Garth Brooks?" That's what Google presented in, People Also Asked, when I did a search for "Garth Brooks." Have people stopped liking Garth Brooks? That's not the point. This is an example of a bigger problem with AI: how a question can contain assumption that have not been established. This is about how AI can give us facts and opinions that make assumptions appear to be true. Learn how changing the way you ask AI a question will produce a more useful and accurate answer.

Can two people have the same experience while assigning completely different meanings to it?

This conversation uses the idea of "friends with benefits" as a case study to explore a broader question: How do expectations, assumptions, and personal meaning shape the way we experience the same situation differently?

In this episode we explore:

• emotional attachment and casual relationships
• why people interpret intimacy differently
• hidden expectations and unspoken assumptions
• emotional asymmetry in relationships
• biology, psychology, and attachment
• the difference between physical intimacy and emotional connection
• why ambiguity creates confusion and hurt
• communication and mismatched objectives

A recurring theme throughout the discussion is that people often assume shared understanding when, internally, they may be operating from completely different emotional frameworks.

This episode originally aired on a previous relationship-focused podcast project. What interests me is the broader pattern of human behavior, interpretation, emotional decision-making, and how people navigate ambiguity in relationships.

What does democratic socialism mean?

I wasn't sure.

Rather than deciding whether it was good or bad, I began investigating how the term is defined by different sources.

This episode explores why political conversations become unproductive before they even begin—not because people disagree - rather they are using the same words to mean different things.

• Why definitions matter before debating ideas
• The differences between Wikipedia, political organizations, and AI-generated summaries
• How labels shape our thinking before we evaluate evidence
• Why broad summaries don't answer the question we're asking
• A better framework for evaluating political ideas, one proposal at a time

The episode isn't about democratic socialism.

It's about how to think clearly when we encounter political language, unfamiliar terminology, or competing definitions. Before asking whether you agree with an idea, first ask:

"What do you mean?"

Data centers have become one of the most controversial pieces of modern infrastructure. Most discussions focus on artificial intelligence, electricity demand, water consumption, and carbon emissions. But are those the right questions?

In this episode, attorney and civil liberties advocate Jennifer Pelton shares a different perspective. She argues that surveillance—not AI—is a major force behind the rapid expansion of data centers and discusses why she believes important issues are being overlooked.

We explore:

• what data centers are and how they've evolved
• the different types of data centers and why those distinctions matter
• surveillance, privacy, and constitutional rights
• the idea of "controlled opposition"
• environmental and community impacts, including mining, noise, infrasound, and electrical infrastructure
• common assumptions about AI, renewable energy, and cloud storage
• what citizens can do if a data center is proposed in their community

Throughout the conversation, one question keeps returning:

How do we know that?

Whether you agree, disagree, or remain undecided, this episode is an invitation to examine evidence, question assumptions, and think more deeply about one of the fastest-growing technologies shaping modern society.

SHOW NOTES

Jennifer L. Pelton, Esq. NY & CT Attorney

Substack: https://substack.com/@jennesq LinkedIn: https://www.linkedin.com/in/jennifer-l-pelton-esq/ Linktree": https://linktr.ee/jenniferpelton

Why do so many people investigating the same topic end up reaching similar conclusions?

Is it because the evidence points in one direction? Or is it because recommendation algorithms reinforce the first question they asked?

In this episode, I explore how social media algorithms shape the information we encounter—not by deciding what's true or false - by giving you more of what you're interested in. Using examples from AI-generated music, data centers, and water consumption, I examine how the first article, video, or headline we encounter can quietly frame the rest of our investigation.

The challenge isn't misinformation. It's that we become experts within the original frame of a problem without stepping outside it to ask a different question that may lead somewhere entirely different.

Sometimes the most important discovery isn't finding a better answer - it's realizing you started with asking the wrong question.

This conversation started as a discussion about "taking a break" in a relationship. Underneath is a broader question about compatibility, emotional pressure, communication, boundaries, and how people respond when relationships begin to feel psychologically overwhelming.

When someone asks for space, what are they actually communicating? Is it a temporary reset, avoidance, incompatibility, emotional overload, or the beginning of the end?

In this episode we explore:

• why some relationships begin to feel smothering
• how people communicate discomfort indirectly
• the difference between needing space and wanting out
• emotional pressure and relationship dynamics
• compatibility under stress
• whether distance clarifies or accelerates underlying problems

This episode originally aired on a previous relationship-focused podcast project. What interests me now is the broader pattern of human behavior, communication, interpretation, and decision-making under emotional uncertainty.

AI companies have been accused of training music-generation models on copyrighted songs without permission. Lawsuits followed. Licensing deals emerged. The debate became about copyright and compensation. While investigating the issue, I found myself asking a different question:

How did the music actually get into the training system?

That question led me into datasets, metadata, YouTube links, and an under explored part of the public discussion—the pipeline between publicly available music and AI model training.

In this episode, I explore why datasets are not the same as audio collections and why understanding how a system works is as important as deciding what should happen after the fact.

This isn't an argument for or against AI companies or artists. It's an exploration of problem definition, assumptions, and why understanding the mechanism leads to better questions—and better solutions.

Jason Earle says he created the "got mold? Test Kit" to make air sampling accessible and affordable. In this conversation, we explore what problem air sampling is intended to solve, the role of mold inspectors, and the difference between testing for mold and mold remediation. A recurring theme throughout this podcast is:

Before choosing a solution, start by asking what problem you're trying to solve.

An exploration of how the questions we ask—and the models we build—can influence narratives that shape technology, investment, and public policy.

A widely cited paper estimated the water footprint of AI models. The results spread rapidly through news stories, social media, and public debate. But what question was the researchers actually trying to answer?

I explore the assumptions behind lifecycle water accounting and how assumptions are shaping conversations about AI, data centers, and proposals for space-based computing.