Breno's Slides – Interpolation versus Extrapolation

Interpolation versus Extrapolation

Sci-Fi Classic

Original RoboCop (1987) audience and critic score card

Blend Action Thriller

RoboCop remake (2014) audience and critic score card

AI is sea level

A shared baseline, not a summit.

AI raises the floorWeak work moves toward average.

Over-reliance lowers the ceilingDistinctive work moves toward average.

shared-model average

Use AI to reach the surface.
Bring your own judgment above it.

Original tweet:

“If you are relying on AI to write for you and to generate ideas for you, you are putting yourself at sea level” Silicon Valley’s go-to comms strategist Lulu on why templates and AI writing rob you of being better “When you use a template, you are giving up on the possibility that you could be better. And I say this sometimes about AI writing, AI is a wonderful tool, I use it all the time” “But if you are relying on AI to write for you, you’re putting yourself at the level of anyone else in the population who wants to pay fractions of a cent for a token to get similar output from the same shared brain. So if you are far below average, you should do that, because average is better than you. But if you believe you can be even a little bit better than average, you would be robbing yourself” “Ralph Waldo Emerson calls it a form of suicide, because in imitating others, which is what a template is, you are killing your own ideas. What are we if not a unique combination of beliefs, ideas, feelings?” “If you take that and you mute it, you are killing a really sacred part of yourself. So using templates is robbing yourself of the possibility that you could ever be better than the last thing, and that’s a shame”

[Sources]

Original post by @clipsofcrypto on X: https://x.com/clipsofcrypto/status/2088827366491762895
User-provided clipping: C:\Users\breno\OneDrive\hub\Clippings\Post by @clipsofcrypto on X.md

Who is using whom?

The same tools can create opposite relationships.

Reverse centaur AI you The system sets the route, pace, and task. ride-hailing · delivery · picking

Centaur You AI You set the goal, constraints, and judgment. research · analysis · creation

A warehouse order picker with a horse head, hypnotized by the light of a smartphone

A classic centaur with a human upper body and horse body

The goal: keep the human in the head.

The metaphor is about agency, not anatomy:

In a reverse-centaur system, the machine decomposes and directs the work while the person supplies perception, movement, and accountability. App-dispatched ride-hailing and delivery work make the pattern intuitive.
In a centaur system, the person chooses the objective and exercises judgment while the machine supplies speed, search, drafting, and scale.
The practical question is not simply whether AI is present, but who sets the goal and who serves whom.

Attribution note: the earliest public use of “reverse centaur” in this automation sense located for this slide is Cory Doctorow’s 17 February 2021 essay. Garry Kasparov pioneered human-computer “Advanced Chess” in 1998; chess players later adopted the centaur label. Kasparov should not be credited with coining the word itself, whose human-machine metaphor has earlier precedents.

[Sources]

Cory Doctorow, “Reverse centaurs and the failure of AI” (17 February 2021): https://pluralistic.net/2021/02/17/reverse-centaur/
ChessBase, Garry Kasparov on the first public Advanced Chess event in June 1998 and the later “centaur” label: https://en.chessbase.com/post/how-technology-is-changing-our-minds/25
Wordorigins, “centaur, reverse centaur” (etymology and attribution caveat): https://wordorigins-org.ghost.io/centaur-reverse-centaur-free-copy/
Matched illustrations generated for this deck with OpenAI image generation and isolated to transparent backgrounds: assets/reverse-centaur-ai-driven-transparent.png and assets/centaur-transparent.png.

Perspective comes from a life lived

A person holding a mask before layered urban and digital imagery.

The human edge

AI can imitate patterns.

Your perspective comes from what you have lived.

What you notice and what you reject depend on who you are.

Experience → taste → judgment

This slide separates pattern imitation from perspective. Perspective means what a person notices, values, and rejects based on experience, identity, relationships, and consequences.

The image carries the ambiguity. The mask suggests an output that can look human, while the eyes and layered urban and digital scenes point to the life behind judgment. Let the audience look before explaining the visual.

Do not use this slide to claim that AI has no creative capacity or that human judgment is always correct. The narrower claim is that lived context shapes selection, taste, and accountability.

Suggested talk track: “AI can reproduce patterns from vast amounts of human work. But it has not lived your constraints, relationships, failures, or consequences. Those experiences shape what you notice, what you value, and what you are willing to reject.”

[Sources]

Breno Alves Beirigo, “Learning in an AI Era” (11 November 2025): https://brenobeirigo.github.io/articles/article_creativity-and-ai.html
Image as published with that article: https://brenobeirigo.github.io/articles/img-programming-ai/mask_cellphone_word_self_human.jpg
The displayed image is an AI-assisted restoration of the article image, created with OpenAI image generation to reduce compression artifacts and improve projection clarity.
Provenance note: The article page does not identify the image’s original creator or license. Confirm both before public redistribution.

The Spartan rule of AI

Different tools. The same test: judgment.

Ancient accounts of Sparta

Steal food

The practice trained resourcefulness.

A Spartan youth carrying a loaf and figs, illustrated in the style of Attic pottery.

Punished when caught
The failure was not stealing.

It was stealing badly.

AI writing

Steal ideas

AI makes borrowed patterns effortless.

An author stealing speech-bubble ideas while saying “It’s not stealing—it’s transformative ideation,” illustrated in the style of Attic pottery.

Penalized when the borrowed voice shows
The failure is not assistance.

It is losing authorship.

For sure, use AI. You still own the result.

Caught means the work shows that you did not verify, understand, or control the output.

Historical basis: Xenophon, Constitution of the Lacedaemonians 2.5–2.8; Plutarch, Lycurgus 17.3–4.

Provenance summary. This slide carefully paraphrases two ancient literary accounts. Xenophon says Spartan boys were deliberately kept somewhat hungry, allowed to relieve their hunger by stealing, and beaten when caught because they had performed the assigned skill badly. Plutarch later repeats the food-stealing practice and describes a caught boy as a careless and unskilful thief. Since these are literary accounts—especially Plutarch, writing centuries later—the slide says “Ancient accounts of Sparta” rather than presenting the anecdote as directly verified social history.

The analogy does not encourage concealing unauthorized AI use. AI use is permitted. “Caught” means the submitted work itself exposes careless tool use, such as invented citations, unsupported claims, generic or internally inconsistent prose, undetected errors, or reasoning the student cannot explain. The student remains responsible for the final work.

The right-hand figure exaggerates generic AI imitation as “stealing ideas.” Its speech bubble says, “It’s not stealing—it’s transformative ideation.” This is a teaching metaphor, not a legal claim about copyright or model training.

[Sources]

Xenophon, Constitution of the Lacedaemonians 2.5–2.8 (Xenophon 1925): https://topostext.org/work/798
Plutarch, Life of Lycurgus 17.3–4 (Plutarch 1914): https://penelope.uchicago.edu/Thayer/E/Roman/Texts/Plutarch/Lives/Lycurgus*.html
Jean Ducat, Spartan Education: Youth and Society in the Classical Period (Ducat 2006).
Nigel Kennell, Spartans: A New History (Kennell 2010).
Both visible illustrations are original AI-generated assets created for this deck. They borrow Attic pottery’s visual language but are not archaeological reconstructions.

Iron sharpens iron

Proverbs 27:17

AI alone

Safe practice

Follows your framing

Describes culture

AI + people

Real feedback

Challenges your framing

Exposes you to culture

AI can help you practice. People sharpen your thinking.

The title comes from Proverbs 27:17. The Hebrew text literally describes iron sharpening iron and a person sharpening the “face” of a friend. The NET translation notes interpret this as interaction and constructive criticism sharpening character (NET Bible, n.d.). A later rabbinic reading applies the proverb to two scholars sharpening one another through joint study (Sefaria, n.d.).

AI provides low-risk explanation, rehearsal, and iteration. That benefit depends on how learners use it. In a field experiment, unrestricted GPT access improved assisted practice but reduced subsequent unassisted performance; tutor-style guardrails largely removed the harm (Bastani et al. 2025). Human interaction adds disagreement, accountability, tacit norms, and participation in a living culture. Research on professional communication training also warns that apparently cooperative AI interactions can entrench simplified cultural narratives (Dai et al. 2025).

Suggested talk track: “AI can be a patient rehearsal partner. But a model cannot induct you into a living culture by itself. Bring your ideas to people who can disagree, correct, and surprise you. That friction is what sharpens them.”

[Sources]

Proverbs 27:17 and NET translation notes (NET Bible, n.d.).
Babylonian Talmud, Ta’anit 7a:12 (Sefaria, n.d.).
Bastani et al., “Generative AI Can Harm Learning” (Bastani et al. 2025).
Dai et al., work on critical interactional competence and generative AI (Dai et al. 2025).

Leverage moves through programmers

A provocation about who gains power from AI

AI won’t replace programmers, but rather make it easier for programmers to replace everyone else.

Naval Ravikant

AI capability × programming judgment → more work becomes programmable

Naval Ravikant posted the sentence on X on 3 January 2025. The visible wording is verbatim, with straight quotation marks to match this deck’s typography. In a later discussion, Naval explained the claim as a statement about leverage: a programmer using a fleet of AI systems can create and reshape software much faster than before.

Treat the line as a provocation, not as an empirical forecast that every non-programmer will lose a job. “Replace everyone else” compresses several possible changes: automating tasks, redesigning workflows, shrinking some roles, and allowing small teams to challenge established organizations. The pace and effect vary by occupation, institution, regulation, and access to technology.

The durable advantage is not syntax alone. Here, “programmer” means someone who can translate an objective into a system, reason through abstractions, evaluate outputs, detect failure modes, and repair what the model gets wrong. Domain expertise, judgment, safeguards, and accountability remain essential.

Suggested talk track: “The provocative part is not that programmers are safe. It is that AI makes software creation more powerful. The people who can turn ambiguous goals into testable systems gain leverage over more kinds of work. The practical response is to learn how systems work, then pair that knowledge with domain expertise and responsibility.”

[Sources]

Naval Ravikant, X post (3 January 2025): https://x.com/naval/status/1875297712993964231
Naval Ravikant and Nivi, “A Motorcycle for the Mind,” discussion of the post and its intended meaning: https://nav.al/ai

AI can create a Matthew effect

Experts use the same AI to frame, verify, and compound better decisions.

Different starting expertise

Expert judgment
Novice judgment

Frames the real problem
Tests against the domain
Builds better next questions

same capability

Delegates the framing
Accepts plausible output
Builds shallow fluency

Cumulative advantage

“For unto every one that hath shall be given, and he shall have abundance…” Matthew 25:29 (KJV)

The visible slide presents the AI-specific Matthew effect as a possible expertise multiplier. An expert can use the same model to frame a more valuable problem, detect subtle failures, verify evidence, and convert one answer into a better next question. Repeated across tasks, those advantages can compound. The provenance and limits belong here, not on the canvas.

The name comes from Matthew 25:29 in the King James Version. The visible quote is intentionally excerpted and ends with an ellipsis; the complete verse is linked in Sources. The verse occurs near the end of the Parable of the Talents. It supplies the language for the name; it is not itself a sociological theory or an endorsement of inequality.

Robert K. Merton introduced “the Matthew effect in science” in a 1968 Science paper. He described how already eminent scientists can receive disproportionately more recognition for comparable contributions. Harriet Zuckerman’s interviews with Nobel laureates formed a major part of the paper’s empirical foundation. Merton later stated that the 1968 paper relied so extensively on Zuckerman’s interviews and materials that it should have appeared under joint authorship.

Merton’s 1988 sequel develops the broader mechanism as cumulative advantage. Early differences in capacity, position, resources, and opportunity can lead to further differences. Recognition unlocks resources and opportunities, which can widen later gaps unless countervailing processes intervene.

The AI transfer is not Merton’s original claim. In open-ended work, expertise can improve problem selection, framing, constraints, failure anticipation, evaluation, and verification. The expert path also feeds each result into a better next question. The novice path shows a different risk: delegating the framing and accepting plausible output can create fluency without durable understanding.

AI-specific research supports parts of this mechanism. Sun and colleagues find that employees with stronger metacognitive strategies gain more creative benefit from generative AI. Prather and colleagues observe that successful novice programmers can reject unsuitable AI suggestions, while students with pre-existing metacognitive difficulties can develop an illusion of competence. A longitudinal study of supervised medical students describes a Matthew-effect pathway in which prior technological experience and mastery orientation strengthen the relationship among AI participation, AI literacy, and critical thinking.

A stronger claim about cumulative divergence requires a feedback mechanism. For example, an output may create better opportunities, resources, or positioning for the next task. A 24-week study of music students reports steeper AI-supported growth among students who began with stronger skills and interprets this as a Matthew effect. That result is domain-specific and quasi-experimental. The visible slide therefore claims different value from the same AI, not that every operator gap must grow over time.

The effect is contingent. AI can narrow gaps on bounded tasks. Brynjolfsson, Li, and Raymond report larger productivity gains for less-experienced and lower-skilled customer-support agents. Dell’Acqua and colleagues also find larger benefits for lower performers on tasks inside the AI capability frontier, while AI can reduce correctness outside that frontier. Do not claim that experts always benefit more.

“Garbage in, garbage out” has an independent computing provenance. Its earliest listed printed attestation is E. E. Blanche’s July 1957 article in Traffic Quarterly. Blanche reports that his personnel used the acronym GIGO; this documents workplace usage rather than identifying an individual coiner. The individual originator remains unknown.

Here, “input” means more than the prompt. It includes the goal, task framing, source material, context, constraints, tests, selection criteria, verification, and willingness to reject an attractive answer. Model training, product design, tool access, and safeguards also affect the outcome.

Suggested talk track: “Give the same AI to a novice and an expert. The novice may ask for an answer. The expert frames the real problem, recognizes what is missing, tests the result against the domain, and asks a better next question. The model is shared, but the surrounding expertise is not. Across repeated work, AI can compound that advantage. This is not inevitable: on bounded tasks, AI can also help less-experienced workers catch up.”

[Sources]

The Holy Bible, King James Version, Matthew 25:29: https://www.kingjamesbibleonline.org/Matthew-25-29/
Robert K. Merton, “The Matthew Effect in Science,” Science 159(3810), 1968, pp. 56–63: https://doi.org/10.1126/science.159.3810.56
Robert K. Merton, “The Matthew Effect in Science, II,” Isis 79(4), 1988, pp. 606–623: https://doi.org/10.1086/354848
Harriet Zuckerman, Nobel Laureates: Sociological Studies of Scientific Collaboration, PhD dissertation, Columbia University, 1965: https://garfield.library.upenn.edu/merton/matthewii.pdf
Shuhua Sun, Zhuyi Angelina Li, Maw-Der Foo, Jing Zhou, and Jackson G. Lu, “How and for whom using generative AI affects creativity: A field experiment,” Journal of Applied Psychology 110(12), 2025, pp. 1561–1573: https://doi.org/10.1037/apl0001296
James Prather et al., “The Widening Gap: The Benefits and Harms of Generative AI for Novice Programmers,” ICER 2024: https://arxiv.org/abs/2405.17739
“AI literacy mediates AI-assisted diagnosis participation and critical thinking among medical students under supervision,” npj Digital Medicine, 2026: https://www.nature.com/articles/s41746-026-02521-9
“Long-term impact of ChatGPT-4 on music students’ creative and analytical skills: A follow-up study,” 2026: https://www.sciencedirect.com/science/article/pii/S187118712600218X
Erik Brynjolfsson, Danielle Li, and Lindsey Raymond, “Generative AI at Work,” Quarterly Journal of Economics 140(2), 2025, pp. 889–942: https://doi.org/10.1093/qje/qjae044
Fabrizio Dell’Acqua et al., “Navigating the Jagged Technological Frontier,” Organization Science 37(2), 2026, pp. 403–423: https://doi.org/10.1287/orsc.2025.21838
E. E. Blanche, “Applying New Electronic Computers to Traffic and Highway Problems,” Traffic Quarterly 11(3), 1957, pp. 406–416: https://trid.trb.org/View/690083
Charles Clay Doyle, Wolfgang Mieder, and Fred R. Shapiro, The Dictionary of Modern Proverbs, Yale University Press, 2012: https://yalebooks.yale.edu/book/9780300136029/the-dictionary-of-modern-proverbs/

Don’t build your work on rented land

Digital sovereignty means the system survives when a supplier changes.

Digital feudalism One company owns the stack

interface agents + automations memory + knowledge model

They change the rules
price ↑ · API shifts · product sunset

Your workflow breaks.

Digital sovereignty You own the system

open files + data your knowledge + workflows portable interfaces model A model B

A supplier changes
swap the model, preserve the system

Your workflow continues.

Own the system. Rent the intelligence.

This slide uses “digital feudalism” as an intentionally sharp operational metaphor: dependency grows when one provider owns the model, interface, memory, automations, data formats, and distribution. The issue is not using a large platform; it is allowing the cost of exit to become larger than the value of choice.

The more resilient pattern is architectural sovereignty. Keep durable knowledge in formats and stores you control; keep prompts, workflows, and automations outside any single model where practical; use documented interfaces; and treat the model as a replaceable component. This does not require avoiding cloud or proprietary tools. It means maintaining a credible exit path.

Research context:

The UK Competition and Markets Authority found that technical and commercial barriers can make it difficult for cloud customers to switch or use multiple providers, locking them into earlier choices that may no longer fit their needs.
The EU Data Act explicitly addresses cloud switching barriers such as data-egress charges, lengthy procedures, and limited interoperability. It requires open interfaces and machine-readable exports for relevant services, with switching charges prohibited from 12 January 2027.
NIST guidance identifies portability, standardized interfaces, and standardized data formats as safeguards against vendor lock-in.
“Datafeudalism” and “technofeudalism” are debated political-economy concepts. The literature uses the feudal metaphor to describe dependence on privately owned platforms and datasets; here, the metaphor is narrowed to a practical design warning rather than presented as a settled economic diagnosis.

Suggested talk track: “Convenience is not sovereignty. If your model disappears tomorrow, what survives? Your notes, sources, prompts, automations, and relationships should. The model should be a socket—not the house.”

[Sources]

UK Competition and Markets Authority, “CMA independent inquiry group publishes provisional findings in cloud services market investigation” (28 January 2025): https://www.gov.uk/government/news/cma-independent-inquiry-group-publishes-provisional-findings-in-cloud-services-market-investigation
European Commission, “Data Act explained,” Chapter VI on switching between data-processing services: https://digital-strategy.ec.europa.eu/en/factpages/data-act-explained
Regulation (EU) 2023/2854 (Data Act), Article 29 on withdrawal of switching charges: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32023R2854
NIST Special Publication 800-146, “Cloud Computing Synopsis and Recommendations,” sections on portability and interoperability: https://nvlpubs.nist.gov/nistpubs/legacy/sp/nistspecialpublication800-146.pdf
Galindo, L. (2024), “Datafeudalism: The Domination of Modern Societies by Big Tech Companies,” Philosophy & Technology 37:121: https://link.springer.com/article/10.1007/s13347-024-00777-1

References

Bastani, Hamsa, Osbert Bastani, Alp Sungu, Haosen Ge, Ozge Kabakci, and Rei Mariman. 2025. “Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics.” Proceedings of the National Academy of Sciences 122 (26). https://doi.org/10.1073/pnas.2422633122.

ChessBase. n.d. “How Technology Is Changing Our Minds.” ChessBase. Accessed August 26, 2026. https://en.chessbase.com/post/how-technology-is-changing-our-minds/25.

@clipsofcrypto. n.d. “Post on AI Writing, Templates, and ‘Sea Level’.” X. Accessed August 26, 2026. https://x.com/clipsofcrypto/status/2088827366491762895.

Competition and Markets Authority. 2025. “CMA Independent Inquiry Group Publishes Provisional Findings in Cloud Services Market Investigation.” January 28. https://www.gov.uk/government/news/cma-independent-inquiry-group-publishes-provisional-findings-in-cloud-services-market-investigation.

Dai, David Wei, Hua Zhu, and Guanliang Chen. 2025. “How Does Interaction with LLM-Powered Chatbots Shape Human Understanding of Culture? The Need for Critical Interactional Competence (CritIC).” Annual Review of Applied Linguistics 45: 28–49. https://doi.org/10.1017/S0267190525000054.

Doctorow, Cory. 2021. “Reverse Centaurs and the Failure of AI.” February 17. https://pluralistic.net/2021/02/17/reverse-centaur/.

Ducat, Jean. 2006. Spartan Education: Youth and Society in the Classical Period. Translated by Emma Stafford, Pamela-Jane Shaw, and Anton Powell. Classical Press of Wales.

European Commission. n.d. “Data Act Explained.” European Commission. Accessed August 26, 2026. https://digital-strategy.ec.europa.eu/en/factpages/data-act-explained.

European Union. 2023. Regulation (EU) 2023/2854 on Harmonised Rules on Fair Access to and Use of Data (Data Act). https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32023R2854.

Galindo, L. 2024. “Datafeudalism: The Domination of Modern Societies by Big Tech Companies.” Philosophy & Technology 37. https://doi.org/10.1007/s13347-024-00777-1.

Kennell, Nigel M. 2010. Spartans: A New History. Wiley-Blackwell.

National Institute of Standards and Technology. n.d. Cloud Computing Synopsis and Recommendations. Special Publication 800-146. National Institute of Standards and Technology. Accessed August 26, 2026. https://nvlpubs.nist.gov/nistpubs/legacy/sp/nistspecialpublication800-146.pdf.

NET Bible. n.d. “Proverbs 27:17.” Bible.org. Accessed August 26, 2026. https://classic.net.bible.org/passage.php?passage=Pro+27:17.

Plutarch. 1914. “The Life of Lycurgus.” In Plutarch’s Lives, Volume i, translated by Bernadotte Perrin. Loeb Classical Library 46. Harvard University Press. https://penelope.uchicago.edu/Thayer/E/Roman/Texts/Plutarch/Lives/Lycurgus*.html.

Sefaria. n.d. “Taanit 7a:12.” Sefaria Library. Accessed August 26, 2026. https://www.sefaria.org/Taanit.7a.12.

Wordorigins.org. n.d. “Centaur, Reverse Centaur.” Wordorigins.org. Accessed August 26, 2026. https://wordorigins-org.ghost.io/centaur-reverse-centaur-free-copy/.

Xenophon. 1925. “Constitution of the Lacedaemonians.” In Xenophon: Scripta Minora, translated by Edgar Cardew Marchant. Loeb Classical Library 183. Harvard University Press. https://topostext.org/work/798.