# What do we do after we beat the Ender Dragon?

By Arya Garg

Canonical source: https://aryagarg23.com/writing/after-the-ender-dragon

Published: 2026-09-22

Register: Speculative

Tags: THOUGHT_EXERCISE

Section: [Futures](https://aryagarg23.com/futures)

## Main

### The dragon as a tutorial

On September 21, 2026, OpenAI announced[^after-the-ender-dragon-main-1] that an internal model had “resolved more than 100 long-standing open problems across most areas of mathematics.”

I thought about Minecraft.

You can spend a long time preparing to defeat the Ender Dragon. You find materials, make equipment and learn to survive in unfamiliar places. The dragon gives that preparation a goal. Defeat it, and the credits can roll[^after-the-ender-dragon-main-2].

Then the world is still there.

I think of the route to the dragon as a tutorial. The things you learned along the way can also help you build a railway or a city with friends.

Minecraft actually starts after you have beaten it.

### Creative mode

But suppose an automated player does all of that for me. When I come back the dragon is dead, and I still don’t know anything about Minecraft as a game. Completing the objective didn’t teach me what else I could do there.

A hundred reported solutions don’t finish mathematics. But take automation to its extreme: suppose machines become excellent at finding answers, explaining them and suggesting worthwhile questions.

I build research tools and spend weekends at hackathons, and I’d gladly automate the setup that keeps me from trying an idea. Minecraft already has a version of this. Creative mode[^after-the-ender-dragon-main-2] hands you the resources, so you learn to build by building.

### The instrument in front of me

Suppose I need a bass sound for a track. An assistant could generate a sample, or help me build a small synthesizer. With the synth, I turn a control, hear something unexpected and ask what changed. I’m learning to shape the sound without having to write the audio software.

A sample might be all I need to finish the track and go to bed. But even a generated sample could make me curious about how it was made, and then the evening goes to experimenting.

I asked for a sound; I did not yet know I would enjoy making sounds.

I can’t always describe that interest in advance. Hidi and Renninger[^after-the-ender-dragon-main-5] describe a passing interest growing through further engagement and support. A tool could get me far enough into sound design to find out whether I enjoy it, without deciding that for me or turning every request into a lesson.

AI tutoring studies separate getting the right answer with help from doing the next problem alone. Students with a less restricted tool[^after-the-ender-dragon-main-4] did better on practice and worse on the exam without it. A carefully designed tutor[^after-the-ender-dragon-main-4] beat the classroom lessons it was compared with. With the synth, I want to get better at shaping sounds, and continued help with the software lets me do that.

### Widening the cone

Imagine a cone containing the questions I can realistically investigate with my current knowledge, resources and time. A useful tool could widen it across fields and deepen it within an unfamiliar one.

Some questions mean comparing thousands of papers, more than I could read in a useful timeframe. A system that organizes them and keeps their sources attached lets me follow a connection without holding the whole archive in my head. Don Swanson called what’s waiting there undiscovered public knowledge[^after-the-ender-dragon-main-6]: findings published separately that support a discovery once someone connects them. The connections still need checking.

[Interactive figure: Wider across fields. Deeper into each]

Something too difficult to try can become an interest once I have a way into it. So automation isn’t a machine working through a fixed list of things humans want. Completing one request can give me a reason to make another I could not have thought of beforehand.

Daniel Litt[^after-the-ender-dragon-main-3] distinguishes producing mathematical results from human understanding of them. A proof can be published and still be new to me. An assistant helping me understand it could hand me a question I want to investigate.

### Chosen constraints

Not every difficulty is in the way.

On the drive to a trailhead, a bypass around the bends leaves more time for the walk. On a Group B rally stage, the bends are what the driver and co-driver came for; remove them and they are competing at something else. Bernard Suits[^after-the-ender-dragon-main-7] calls these the traveler and the racer.

[Interactive figure: Same destination. Different reasons]

Some friction is part of an activity I chose, and some just gets in my way. I can enjoy experimenting with a recipe and still leave the whisking to a mixer.

Nor does every activity need to teach me something. I enjoy music more with friends, even a recording I could play alone; their reactions are part of it, and studies of live music[^after-the-ender-dragon-main-8] link that shared feeling to enjoyment and meaning. I would call the playful part of this whimsy: dancing at a rave, a ridiculous game made up with friends, no need to get better at either. I do not want freedom from work to come with an obligation to spend every spare hour improving ourselves.

### The world we actually have

Automating someone’s job does not mean they can afford to stop working. Someone searching for replacement income is in a different position from someone whose hours fell without a pay cut.

The same goes for learning at work. If software takes over a report a junior employee used to prepare, their employer could have them dig into the results with an experienced colleague. Or the employer could decide the junior position is no longer needed.

There could be more projects worth trying than ever while someone has less money or time to try them. “People can still do it for fun” does not answer that.

I would love more time to explore. Someone else might mainly want fewer demands on their day. Pope Leo XIV’s Magnifica Humanitas[^after-the-ender-dragon-main-9] rejects measuring a person’s dignity by their abilities or achievements. Nobody should have to prove they are still useful by replacing their old job with some impressive new pursuit.

### Bounded infinities

Between two distinct rational numbers lies another.[^after-the-ender-dragon-main-10] Take the midpoint, then subdivide again; there is always another number, and the interval never grows. Learning the basic rules of a subject works like that for me. It lets me look closer rather than meaning I’m done with it.

[Interactive figure: Still room between]

Jevons[^after-the-ender-dragon-main-11] argued that burning coal more efficiently could increase how much of it gets burned. Inquiry might go the same way as it gets cheaper: an afternoon instead of a month makes new questions worth asking. Who decides how the saved time gets spent is still open. I could attempt something new, or my employer could expect more of the same work.

### After the dragon

After the dragon, I might build a railway that takes longer to construct than walking to its destination ever would. A machine could lay the track while a friend and I design the station. We might get carried away and start building a town around it instead. By the end of the afternoon, reaching the original destination could be the least interesting part.

When does getting distracted become finding something worth doing?

[^after-the-ender-dragon-main-1]: OpenAI, “Advisory Group on Mathematics and Artificial Intelligence”, September 21, 2026. [Article](https://openai.com/index/advisory-group-on-mathematics-and-ai/) · [Original X post](https://x.com/OpenAI/status/2102093145051943229)

[^after-the-ender-dragon-main-2]: Mojang, “What Is Minecraft?”, June 5, 2023; Duncan Geere, “End City”, July 18, 2025. [What Is Minecraft?](https://www.minecraft.net/en-us/article/what-minecraft) · [End City](https://www.minecraft.net/en-us/article/end-city)

[^after-the-ender-dragon-main-3]: Daniel Litt, “A Beginning for Mathematics”, September 13, 2026. [A Beginning for Mathematics](https://www.daniellitt.com/blog/2026/9/13/a-beginning-for-mathematics/)

[^after-the-ender-dragon-main-4]: Hamsa Bastani and colleagues, “Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics”, PNAS 122(26), e2422633122, June 2025. Greg Kestin and colleagues, “AI Tutoring Outperforms In-Class Active Learning: An RCT Introducing a Novel Research-Based Design in an Authentic Educational Setting”, Scientific Reports 15, 17458, 2025. [PNAS article](https://doi.org/10.1073/pnas.2422633122) · [Scientific Reports article](https://doi.org/10.1038/s41598-025-97652-6)

[^after-the-ender-dragon-main-5]: Suzanne Hidi and K. Ann Renninger, “The Four-Phase Model of Interest Development”, Educational Psychologist 41(2), 111–127, 2006. [Journal article](https://doi.org/10.1207/s15326985ep4102_4) · [ERIC record](https://eric.ed.gov/?id=EJ736298)

[^after-the-ender-dragon-main-6]: Don R. Swanson, “Undiscovered Public Knowledge”, The Library Quarterly 56(2), 103–118, April 1986. [The Library Quarterly](https://www.journals.uchicago.edu/doi/10.1086/601720)

[^after-the-ender-dragon-main-7]: Bernard Suits, The Grasshopper: Games, Life and Utopia, excerpted by the publisher in “Philosophizing About Games”. [Philosophizing About Games](https://broadviewpress.com/philosophizing-about-games/)

[^after-the-ender-dragon-main-8]: Nicole Koefler, Esha Naidu, Shira Gabriel, Veronica Schneider, Gabriela S. Pascuzzi and Elaine Paravati, “Let the Music Play: Live Music Fosters Collective Effervescence and Leads to Lasting Positive Outcomes”, first published online in 2024. [Journal article](https://doi.org/10.1177/01461672241288027) · [Author manuscript/full text hosted by Duke University](https://dukespace.lib.duke.edu/server/api/core/bitstreams/27f9ff43-bad1-4910-b1c5-abdae85f2a38/content)

[^after-the-ender-dragon-main-9]: Pope Leo XIV, Magnifica Humanitas, May 15, 2026, particularly paragraphs 50–51, which distinguish human dignity from abilities, efficiency and achievement. [Magnifica Humanitas](https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html)

[^after-the-ender-dragon-main-10]: Harvey Mudd College Department of Mathematics, “Rationals Dense but Sparse”. [Rationals Dense but Sparse](https://math.hmc.edu/funfacts/rationals-dense-but-sparse/)

[^after-the-ender-dragon-main-11]: William Stanley Jevons, The Coal Question, Chapter VII, “Of the Economy of Fuel”, first published 1865; linked text is the 1866 second edition. [The Coal Question](https://www.econlib.org/library/YPDBooks/Jevons/jvnCQ.html?chapter_num=9)

## Sources and notes

Revised September 22, 2026. Source links and qualifications are retained from the preceding draft; this is an editorial revision, not a new source check. The automation, synthesizer, workplace and post-dragon scenarios are thought experiments. The playthings illustrate distinctions rather than supply evidence.

### 1. OpenAI's mathematics announcement

OpenAI, “Advisory Group on Mathematics and Artificial Intelligence”, September 21, 2026. [Article](https://openai.com/index/advisory-group-on-mathematics-and-ai/) · [Original X post](https://x.com/OpenAI/status/2102093145051943229)

The opening directly quotes the announcement’s aggregate claim. The preceding source record notes that the announcement does not present an itemized verification record for all of the reported results. The essay attributes the aggregate claim to OpenAI; it does not independently certify the solutions. Original advisory-group announcement on X, retained from the earlier source record; it is not the basis for the quotation.

### 2. Minecraft, progression and Creative mode

Mojang, “What Is Minecraft?”, June 5, 2023; Duncan Geere, “End City”, July 18, 2025. [What Is Minecraft?](https://www.minecraft.net/en-us/article/what-minecraft) · [End City](https://www.minecraft.net/en-us/article/end-city)

The first explains sandbox play and the different conditions of Creative and Survival modes; the second explicitly discusses play after defeating the dragon and rolling credits. Treating the dragon progression as a tutorial is the author’s interpretation, not a mandatory sequence. Creative mode is not unlocked by beating the dragon. The essay makes no claim of literal physical or computational infinity.

### 3. Results and human understanding

Daniel Litt, “A Beginning for Mathematics”, September 13, 2026. [A Beginning for Mathematics](https://www.daniellitt.com/blog/2026/9/13/a-beginning-for-mathematics/)

The distinction between mathematical production, understanding and the communities that cultivate it is an acknowledged intellectual foundation. Litt also discusses institutions using the same achievement to signal both progress in a subject and a person’s expertise. This essay extends the question into developing interests and choosing forms of participation; it does not claim to originate the result–understanding distinction.

### 4. Assisted performance and learning

Hamsa Bastani and colleagues, “Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics”, PNAS 122(26), e2422633122, June 2025. Greg Kestin and colleagues, “AI Tutoring Outperforms In-Class Active Learning: An RCT Introducing a Novel Research-Based Design in an Authentic Educational Setting”, Scientific Reports 15, 17458, 2025. [PNAS article](https://doi.org/10.1073/pnas.2422633122) · [Scientific Reports article](https://doi.org/10.1038/s41598-025-97652-6)

Randomized field experiment at a high school in Turkey in the 2023–2024 academic year, using two GPT-4-based tutor configurations. Outcomes distinguish supported practice from a subsequent unassisted exam. The guarded configuration mitigated the loss seen with the less restricted tool; it did not significantly outperform the control on the exam. This is a bounded short-term learning study, not evidence that all AI assistance causes skill loss. The publisher’s subsequent correction concerns an author affiliation. A crossover study involving 194 eligible Harvard undergraduate physics students and two lessons. It reports immediate post-test learning gains under the compared teaching conditions, not an established long-term outcome or a comparison with every form of teaching. It is not a head-to-head replication of the Bastani study.

### 5. How an interest develops

Suzanne Hidi and K. Ann Renninger, “The Four-Phase Model of Interest Development”, Educational Psychologist 41(2), 111–127, 2006. [Journal article](https://doi.org/10.1207/s15326985ep4102_4) · [ERIC record](https://eric.ed.gov/?id=EJ736298)

A proposed model drawing on prior research, distinguishing situational interest from more developed individual interest and discussing repeated engagement and support. It does not test this essay’s AI scenario or guarantee that any particular encounter becomes a lasting interest. Accessible reproduction of the paper.

### 6. Connecting existing knowledge

Don R. Swanson, “Undiscovered Public Knowledge”, The Library Quarterly 56(2), 103–118, April 1986. [The Library Quarterly](https://www.journals.uchicago.edu/doi/10.1086/601720)

The publisher’s abstract describes how independently published fragments can support discovery when retrieved, combined and interpreted, while stressing retrieval’s incompleteness. Used as a precedent for literature-based discovery, not proof of present-day model accuracy or exhaustive retrieval.

### 7. Games and chosen constraints

Bernard Suits, The Grasshopper: Games, Life and Utopia, excerpted by the publisher in “Philosophizing About Games”. [Philosophizing About Games](https://broadviewpress.com/philosophizing-about-games/)

Suits supplies the traveler–racer distinction. The Group B/trailhead contrast and baking example are adaptations developed for this essay. They are not quotations, a reconstruction of a historical rally stage, or an argument that danger and difficulty are inherently beneficial. The source consulted is the publisher’s excerpt, not a claimed reading of the entire book.

### 8. Collective effervescence

Nicole Koefler, Esha Naidu, Shira Gabriel, Veronica Schneider, Gabriela S. Pascuzzi and Elaine Paravati, “Let the Music Play: Live Music Fosters Collective Effervescence and Leads to Lasting Positive Outcomes”, first published online in 2024. [Journal article](https://doi.org/10.1177/01461672241288027) · [Author manuscript/full text hosted by Duke University](https://dukespace.lib.duke.edu/server/api/core/bitstreams/27f9ff43-bad1-4910-b1c5-abdae85f2a38/content)

Four studies with 789 participants. The essay uses the reported relationships between collective effervescence, enjoyment and meaning; it does not infer that friction causes these outcomes, that collective experience is disappearing historically, or that prerecorded and solitary listening have no value. The observation about enjoying music with friends is the author’s personal account, not a finding of these studies. “Whimsy” is his informal term for playful participation, not a substitute definition of collective effervescence.

### 9. The person beyond performance

Pope Leo XIV, Magnifica Humanitas, May 15, 2026, particularly paragraphs 50–51, which distinguish human dignity from abilities, efficiency and achievement. [Magnifica Humanitas](https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html)

Used as a philosophical and theological reference. The argument here does not depend on the encyclical’s separate judgments about AI capabilities or imply agreement with all of its positions.

### 10. Boundedness and subdivision

Harvey Mudd College Department of Mathematics, “Rationals Dense but Sparse”. [Rationals Dense but Sparse](https://math.hmc.edu/funfacts/rationals-dense-but-sparse/)

Between two distinct rationals, their midpoint is another rational. The figure illustrates continued subdivision in a bounded interval. Neither the theorem nor the animation establishes infinitely many meaningful activities, infinite physical information, or unlimited resources.

### 11. Jevons

William Stanley Jevons, The Coal Question, Chapter VII, “Of the Economy of Fuel”, first published 1865; linked text is the 1866 second edition. [The Coal Question](https://www.econlib.org/library/YPDBooks/Jevons/jvnCQ.html?chapter_num=9)

Jevons argues that greater efficiency can expand uses and aggregate demand. The connection to inquiry is an analogy. It is not a forecast that demand must always offset efficiency gains or that any resulting increase in activity improves people’s lives.