qedbot

Breaking·First-hour kit

Ten advances on open problems

D3 checkable OpenAI 1 Aug 2026

Each result is published with a Lean certificate. Three resolve Erdős problems held here, 146, 180 and 183, and the files that prove them were rebuilt here. Read the announcement.

The first hour

What is claimed, and how much can be inspected?

Ten advances on open problems in mathematics and theoretical computer science. Of 10 results announced, 10 are named, 10 written up and 10 published with an artifact anyone can check.

Has anyone independent checked it?

2 of the 3 results held here are machine-checked.

Does the formal statement say what is claimed?

The authors declare how their formal statement matches the claim; no independent statement anchors it.

  • Erdős Problem 146: Formal statement held: FormalConjectures/ErdosProblems/146.lean. F2 declared: The statement corpus cites this proof against its own statement. Read the statement
  • Erdős Problem 180: Formal statement held: FormalConjectures/ErdosProblems/180.lean. F2 declared: The statement corpus cites this proof against its own statement. Read the statement
  • Erdős Problem 183: Formal statement held: FormalConjectures/ErdosProblems/183.lean. F2 declared: The statement corpus cites this proof against its own statement. Read the statement

Who did the work?

Graded A? declared, not graded for who did the mathematics.

  • OpenAI internal model: A? declared, not graded. The page credits an internal model at OpenAI and does not say how it was directed.
  • OpenAI internal model: A? declared, not graded. The page credits an internal model at OpenAI and does not say how it was directed.
  • OpenAI internal model: A? declared, not graded. The page credits an internal model at OpenAI and does not say how it was directed.

Who got there first?

No precursor, concurrent work or dispute is recorded against these results.

What would move the grades

  • Evidence rises from V2 with an independent rebuild, by qed.bot or Palomar.
  • Fidelity rises from F2 with a Comparator check against a statement from a separate corpus.

AI activity

How grades work
OpenAI internal model

1 Aug 2026

Disproved: a connected bipartite 2-degenerate graph H with ex(n; H) ≫ n^(3/2 + c) for some constant c > 0.

full A? V3 F2
Reasoning and sources

Autonomy

The page credits an internal model at OpenAI and does not say how it was directed.

Details

attribution: This was disproved by an internal model at OpenAI, which constructed a connected bipartite $2$-degenerate $H$ such that\[\mathrm{ex}(n;H)\gg n^{3/2+c}\]for some constant $c>0$.

OpenAI internal model

1 Aug 2026

Disproved, by the construction recorded under Erdős problem 575.

full A? V2 F2
Reasoning and sources

Autonomy

The page credits an internal model at OpenAI and does not say how it was directed.

Details

attribution: This was disproved by an internal model at OpenAI - see the remarks in [575] .

OpenAI internal model

1 Aug 2026

Proved that the limit is infinite: R(3; k) ≥ k^((1/3 − o(1))k) for all k ≥ 2.

full A? V3 F2
Reasoning and sources

Autonomy

The page credits an internal model at OpenAI and does not say how it was directed.

Details

attribution: The limit is actually infinite; this was proved by an internal model at OpenAI, which proved that, for all $k\geq 2$,\[R(3;k)\geq k^{(\frac{1}{3}-o(1))k}.\]See also [483] .

Sources

Citing this kit

qed.bot, “Ten advances on open problems: first-hour kit”, https://qed.bot/a/openai-2026-08-01, as of 20 Nov 2026.

How the grades work