ERDŐS PROBLEM № 0 2026 Budapest OPEN Close it →

Construct a statue of Paul Erdős.

In Hungarian, Erdős Pál: family name first, the way the plinth will read.

Required one statue, in Budapest. The proof is constructive, finite, and fundable.

Given
  1. No public statue exists: not in Budapest, not anywhere on Earth.1
  2. Erdős’s problems are now a frontier-AI testbed.
Abstract

Paul Erdős (1913–1996) posed more than a thousand problems and attached about $60,000 in prizes to them.3 Machines have now begun, verifiably, to solve him; the companies around that work are worth about seven trillion dollars.2 The statue count is still zero. We propose the first monument financed by AI laboratories and designed by AI: Erdős, brain open, in Budapest.

§ 1 · Prior work

The examiner

Paul Erdős in Budapest, fall 1992: glasses, light jacket, mid-conversation.
Erdős, Budapest, 1992.

Erdős wrote roughly 1,500 papers with more than 500 collaborators, the most connected body of work in mathematics.4 He owned no house, no car, and one suitcase. He arrived at colleagues’ doors with the line: “My brain is open.”

He priced hard problems in cash because the price carried information.5 Mathematicians still measure their distance from him by Erdős number. The machines now working through his problem list are, in effect, applying for a number of their own.

“A mathematician is a machine for turning coffee into theorems.”
Alfréd Rényi · the most famous thing Erdős never said, but lived.
Erdős vocabulary
epsilon
a child. From ε, an arbitrarily small quantity.
the SF
God, the “Supreme Fascist”: keeper of the best proofs, hider of socks and passports.
The Book
where the SF keeps the perfect proof of every theorem. “You don’t have to believe in God, but you should believe in The Book.”
to preach
to give a mathematics lecture.
noise
music, other than classical.
to die
to stop doing mathematics.
to leave
to die.
~1,500
papers published, the most of any mathematician
1,217
problems catalogued, and counting
500+
direct collaborators
1
suitcase
0
statues, anywhere on Earth

§ 2 · Partial progress

The record

Since 2024, erdosproblems.com has become an unofficial leaderboard for machine mathematics: 1,217 catalogued problems, 548 settled, 200 checked line by line in Lean.6 The AI ledger credits more than twenty systems with primary contributions to 207 problems; 74 are full solutions.7

The labs have moved constants, found proofs, made public mistakes, and closed open questions. They have not funded one statue.

1,217
Erdős problems catalogued
200
settlements checked in Lean
74
AI full solutions logged
0
statues funded by AI labs
The scoreboard
Laboratory Money, 2026 Erdős record Statues funded
OpenAI $852B valuation 201 ledger contributions, 61 full solutions; disproved the 1946 unit distance conjecture
Paper trail
  • Deleted 2025 claim that GPT-5 had solved ten unsolved Erdős problems; Bloom called it a misrepresentation.9
  • GPT-5.4 Pro helped a member of the public settle problem #1196 after prompting and human refinement.13
  • Internal reasoning model disproved Erdős’s 1946 unit distance conjecture.14
0
Google DeepMind ~$185B AI capex Aletheia’s Lean-checked solution to #1051; nine more via AlphaProof Nexus; 700 problems screened
Paper trail
  • AlphaEvolve improved the bound on Erdős’s minimum overlap problem.8
  • Aletheia produced a Lean-checked solution to problem #1051 after screening hundreds of open problems.12
  • AlphaProof Nexus closed nine more open problems the same week Claude reached the unit-distance result.15
0
Anthropic $965B valuation 27 ledger contributions, 8 full solutions; independent unit distance proof
Paper trail
  • Claude reached the unit-distance result independently within a week, with a proof described as more concise.15
0
Harmonic $1.45B valuation Aristotle: 199 ledger entries, 17 full solutions; formalized #728
Paper trail
  • Formalized the first machine solution of an Erdős problem that survived refereeing.11
0
ByteDance ≤$70B AI capex Seed Prover: new proof logged on #379
Paper trail
  • Seed Prover 1.5 logged a new proof on problem #379.10
0
xAI ~$18B in GPUs Grok: one ledger entry, #524 0
Multiscalar two founders, London Fields System: full solution to #690 0
DeepSeek $5.6M a run, famously DeepThink: one attempt logged; it was wrong 0
Alibaba $53B 3-yr AI pledge Qwen: ranked corroborator on ErdosBench, a third-party Erdős-style benchmark 0
Meta AI $1.5T market cap public critique of OpenAI's claim
Paper trail
  • Yann LeCun joined the public criticism after OpenAI’s deleted claim about GPT-5 and Erdős problems.9
0
Ledger / formalization tailAxiom Math / AxiomProver – Logical Intelligence / Aleph Prover – UlamAI – Math Inc / Gauss – Rethlas – ProofCouncil – Principia Math – Trellis – Numina – Archivara – Lean FRO
Benchmark / adjacentMoonshot / Kimi – Z.ai / Zhipu / GLM – MiniMax – Amazon – Apple – Microsoft Research – Mistral – NVIDIA – Thinking Machines – Shanghai AI Lab – Morph Labs – Sakana – Preferred Networks – NAVER – LG AI Research – Upstage – Kakao
No public Erdős tie foundSafe Superintelligence – Reflection AI – Perplexity – Cohere – Baidu / ERNIE – Tencent / Hunyuan – 01.AI / Yi – StepFun – Huawei / Noah / PanGu – SenseTime / SenseNova – iFLYTEK / Spark – Baichuan AI – Xiaomi / MiMo – Samsung Research / Gauss – SK Telecom / A.X – TAIDE – AI2 – Lighthouz AI
Broader model-list labsAI21 Labs – IBM – Writer – G42 – Inflection AI – Magic – Technology Innovation Institute – MBZUAI – Stability AI – Character.ai – Cursor – Cognition – Physical Intelligence – Reka AI – Databricks

§ 3 · Proposed construction

To be designed by AI

The statue should be designed the way the labs already claim progress in mathematics: models propose, people verify.16 Every funding laboratory may put its models on the brief. A jury of mathematicians and sculptors judges blind. Human hands build the winner.

Budapest decides the site. The street outside the Erdős Center would work. So would City Park, where a teenage Erdős traded problems near the statue of Anonymous.17 The important thing is simple: Erdős belongs in the city’s public memory, not only in databases, prizes, and anecdotes.

§ 4 · The prize

What it costs to close

A serious public statue, with plinth, site work, and installation, likely costs $150,000 to $500,000.18 That is not a budget line for a frontier lab; it is a rounding error. The companies on the scoreboard above will spend hundreds of billions of dollars on AI this year; Alphabet’s projected spend alone passes the price of the statue roughly every ninety seconds.19

There is precedent for private monuments to scientists, and for tech money funding public statues.20 There is no precedent for an AI laboratory funding one. That is the point. Anything raised beyond the bronze endows prizes for Erdős problems still open.

ε
Epsilon
any amount

For individuals. A public tranche opens with the domain.

№ 2
Collaborator
$50,000+

Named in the acknowledgments and cast into the plinth in small type.

№ 1
Erdős number 1
$250,000+

Founding funder. Named plainly and permanently on the plinth.

Close Problem № 0.

If you fund, build, or run a laboratory in this testbed, one email starts the statue.

fund@erdosstatue.com

This inbox is read by a person, not a model. The status at the top of this page flips to SOLVED on the day the statue is unveiled in Budapest.

§ 5 · Correspondence

Proposed by

  • Némethi Balázs — family name first, the way the plinth will read.