The B2B Podcast Index
Index
All categories
MarketingSalesSaaSFinanceHROpsLeadershipCustomer SuccessAI & DataProductStartups & FoundersRevOpsEngineering & DevTools
MethodologySubmit
Best of:MarketingSalesSaaSFinanceHROpsLeadershipCustomer SuccessAI & DataProductStartups & FoundersRevOpsEngineering & DevTools
An independent project byFame
SearchBest episodesGuestsInsightsMethodologySubmit a podcast
Index/AI & Data/Unsupervised Learning
Unsupervised Learning artwork

Humans Need Entropy

Unsupervised Learning · 2025-11-16 · 4 min

0:00--:--

Key moments - from our scoring

Substance score

42 / 100

Five dimensions, 20 points each

Insight Density14 / 20
Originality15 / 20
Guest Caliber0 / 20
Specificity & Evidence11 / 20
Conversational Craft2 / 20

Building on Andrej Karpathy's observation that humans tend toward cognitive collapse over time - becoming increasingly predictable as learning rates decrease and behavior ossifies - the host argues that sustained intellectual vitality requires deliberate cultivation of entropy. Drawing on Shannon's information theory, which defines information as the unexpected or novel content in a transmission, the episode explores why predictability emerges as a natural consequence of aging: children surprise us because everything is new, but adults fall into repeating the same stories, jokes, preferences, and thought patterns. The host connects this to observable phenomena like how older people consume the same media repeatedly and their social aperture shrinks. Rather than accept this inevitability, the host details concrete interventions: deep reading of classical rhetoric and writing, studying Christopher Hitchens for non-clichéd language patterns, building AI tools like an 'Increase Entropy' skill in Claude, and developing metrics like 'wows per minute' to measure how often content genuinely surprises. The episode warns that AI delegation amplifies this problem by cementing mediocrity from internet training data, making human effort to maintain novelty increasingly urgent and valuable.

Key takeaways

  • →Humans experience cognitive collapse starting in their mid-to-late 20s as they reduce exploration and increasingly rely on established thought patterns and behaviors.
  • →Information, per Shannon's theory, is defined by novelty and surprise - not repetition - making entropy directly tied to intellectual value and communication impact.
  • →Deliberately consuming sources of freshness (classical literature, rhetorical study, complex thinkers like Hitchens) is a proven method to break out of predictable language and thought patterns.
  • →Metrics like 'wows per minute' can quantify whether your content or communication is actually surprising audiences or merely recycling familiar ideas.
  • →AI systems amplifying training data from mediocre internet sources will make human commitment to novelty-seeking even more economically and socially valuable.

Topics in this episode

Claude AIAndrej KarpathyShannon's information theoryEntropy in cognitionChristopher HitchensFabric AI frameworkRhetorical figures and classical writingInformation theoryCognitive collapseNovelty-seeking behavior

Questions this episode answers

What did Andrej Karpathy mean by humans 'collapsing' as they age?

Karpathy argued that humans become increasingly predictable and repetitive as they age - their learning rate decreases, they revisit the same thoughts, and their behavioral aperture shrinks until they eventually become entirely predictable to those around them.

How does Shannon's information theory explain why novelty matters in communication?

In Shannon's model, information is defined as the part of transmission that isn't repeated or predictable; novelty and surprise are what constitute actual information, while repetition carries no informational value.

What specific tools has the host built to combat cognitive collapse?

The host created an 'Increase Entropy' skill in Claude that generates novel ways of expressing the same thought, and developed an AI prompt in fabric that rates content for 'wows per minute' - measuring how frequently audience members are genuinely surprised.

Why is studying Christopher Hitchens specifically mentioned as a solution?

Hitchens is cited as a source of non-clichéd, sophisticated language patterns and rhetorical figures that help break speakers out of predictable word choices and thought patterns.

How does AI training on internet data worsen the collapse problem?

AI systems learn by processing mediocre, repetitive content from the internet, amplifying predictability and cliché at scale, which makes human effort to seek genuine novelty increasingly important and valuable.

What our scoring noted

Our reviewer’s read on each dimension, with quotes from the episode.

Insight Density

14 / 20

The episode develops a coherent, substantive thesis connecting Karpathy's model-collapse theory to human cognitive decay and the antidote of entropy-seeking. While the core idea is genuinely interesting, the execution relies heavily on one source (Karpathy's Oct 2025 conversation) and the practical applications are somewhat broad. The speaker offers concrete behavioral examples (reading old rhetorical texts, building an 'Increase Entropy' tool, creating a 'wows per minute' metric) but doesn't deeply interrogate trade-offs or failure modes.

humans collapse during the course of their lives. Children have an overfit, yet they will say stuff that will shock you because they're not yet collapsed. But we adults, we end up revisiting the same thoughts.
The main thing I'm asking myself, especially for my own content, is how much of this is new? How often will I'm presenting this? Will the viewer be pleasantly surprised?

Originality

15 / 20

The framing of human cognitive decline as model collapse/overfitting is genuinely fresh and counterintuitive. The linkage to Shannon's information theory and the emphasis on entropy as antidote feels original. However, the underlying concept of cognitive rigidity with age and the need for novelty are not new; the originality resides primarily in the technical framing and the speaker's personal systems (AI tools, metrics) rather than entirely novel insight.

humans collapse during the course of their lives
information as the part of transmission that isn't repeated or noise

Guest Caliber

0 / 20

This is a solo monologue with no guest present. The only referenced practitioner is Andrej Karpathy, discussed secondhand via a conversation the host heard, not interviewed directly. For a B2B learning podcast, the absence of a live expert or practitioner severely limits the opportunity for real-time interrogation, pushback, or detailed case study exploration.

I've had several thoughts on the Carpathian Dwarkesh conversation that took place in late October of 25

Specificity & Evidence

11 / 20

The speaker provides some concrete personal examples (reading rhetoric books, listening to Hitchens, building an 'Increase Entropy' tool, creating a 'wows per minute' metric) but lacks hard data, quantified results, or named case studies of others executing this strategy. There are no metrics on whether these interventions actually work, no timeline for the projects mentioned, and no comparative examples of people who did or didn't arrest cognitive collapse.

I'm reading a lot of old books on writing, like rhetorical figures and stuff like that, to try to get fresh phrases into my mind
I created an AI prompt in fabric that would rate talks, blogs, panels, or whatever for wows per minute

Conversational Craft

2 / 20

This is a scripted solo essay with no host-guest dialogue, follow-up questions, or adversarial pressure. There is no one to challenge the claim that entropy-seeking actually prevents cognitive collapse, no counterargument explored, and no nuance around whether constant novelty-chasing might itself become a new form of collapse (performative variation rather than genuine learning).

I've been terrified of this happening to me.
At least for us humans. The solution seems something like recognize that this is a problem

Conversation analysis

Computed from the transcript - who did the talking, and the words that came up most.

Most-used words

entropy5humans3karpathy3constantly3language3fresh3content3thoughts2conversation2late2whole2human2model2collapse2saying2rate2

Episode notes

How humans and AI models both share the weakness of deterioration without novel inputs. Become a Member: See omnystudio.com/listener for privacy information.

Full transcript

4 min

Transcribed and scored by The B2B Podcast Index.

WEBVTT - Humans Need Entropy I've had several thoughts on the Carpathian Dwarkesh conversation that took place in late October of 25, but the one that keeps haunting me is something Karpathy just kind of casually mentioned before moving on to another topic. I think it might be the biggest idea in the whole conversation. He was talking about human model similarities and he says humans collapse during the course of their lives. Children have an overfit, yet they will say stuff that will shock you because they're not yet collapsed.

But we adults, we end up revisiting the same thoughts. We end up saying more and more the same stuff. The learning rate goes down, the collapse continues to get worse, and then everything deteriorates. End quote.

Since my 20s, I've been terrified of this happening to me. It pierces my soul whenever my partner says things like, I knew you were going to say that. Ouch. Predictable humor or wit isn't another example.

How many older people do you know who tell the same stories and jokes over and over? They watch the same shows. They listen to the same five bands, and then eventually, like 2 or 1, their aperture slowly shrinks until they die. Luckily, Karpathy gives a solution right after we have to find sources of entropy.

When we were kids, everything was entropy because everything was new. So we were constantly changing our preferences, our behaviors, our language and everything. It made us fresh, unpredictable, which is highly related to the concept I'm obsessed with from Shannon's theory of information, which in his model defines information as the part of transmission that isn't repeated or noise. I think about this constantly when I'm giving talks or participating in panels or whatever, or when I'm watching someone else do so.

The main thing I'm asking myself, especially for my own content, is how much of this is new? How often will I'm presenting this? Will the viewer be pleasantly surprised? If the answer is not very often I redo it or I start over.

I'm actively doing a bunch of stuff in addition to pathological reading to maximize entropy in my life. I'm reading a lot of old books on writing, like rhetorical figures and stuff like that, to try to get fresh phrases into my mind. I regularly reread and listen to Christopher Hitchens books and debates. Just having exposure to that level of non cliché language.

And I'm currently building in cloud code a skill called Increase Entropy that incorporates all of this old and fresh language like a particle accelerator. So I can point it at a thought or a piece of content and basically come up with novel ways of saying the same thing. So I give it the way that I would say it in a kind of like just breaks me out of my mold. I even went so far in 2024 to create an AI prompt in fabric that would rate talks, blogs, panels, or whatever for wows per minute, meaning how often a given piece of content surprised the audience.

I mean, this was a problem before AI, and now many are delegating even more and more of their thinking to a system that learns by crunching mediocrity from the internet. I can see things getting way worse in this respect. I guess it's somewhat comforting that this happens to both AI models and to people. It makes the whole thing more human somehow.

And hearing Karpathy say it so plainly was jarring to me in a pleasant way. At least for us humans. The solution seems something like recognize that this is a problem that starts for everyone in there, probably like mid to late 20s, and constantly seek and consume sources of novelty and freshness to maintain young mind.

Related episodes across the Index

Other episodes covering the same guests and topics, from across The B2B Podcast Index.

  • SOPs That Don't Collect DustThe Lean Solutions Podcast · on Claude AI83 / 100
  • The Power of You Encoded Content, with Jodi Kay EdwardsBuilding your LeaderBrand · on Claude AI80 / 100
  • How To Build A Million Dollar Biller. Danny Cahill.The Elite Recruiter Podcast · on Claude AI78 / 100
  • How can agency owners hold themselves accountable?Agency Leadership Podcast · on Claude AI78 / 100
  • 🟡NEW! Amy Porterfield! Webinars, Email, and Events! | Ep. 525Do This, NOT That: Marketing Tips with Jay Schwedelson · on Claude AI78 / 100
  • How AI and Revenue Orchestration Are Reshaping B2B Marketing StrategiesThe B2B Revenue Executive Experience · on Claude AI77 / 100

More from Unsupervised Learning

All episodes →
  • AI Predicts the Text of Answers52 / 100
  • Most Companies Aren't Anywhere Near Ready for AI61 / 100
  • We're All Building a Single Digital Assistant67 / 100
  • Why AI Will Replace Knowledge Workers66 / 100
  • Why I Believe in SOTA Models Over Custom Ones49 / 100
Explore the best B2B AI & Data podcasts →
All Unsupervised Learning episodes →