Skip to content

Research Story

Aaronson-Ambainis conjecture: AI-assisted proofs claim to settle a 2011 question on quantum advantage

According to New Scientist, David Turturean of Physical Superintelligence and colleagues used Anthropic's Fable 5.2 to prove the Aaronson-Ambainis conjecture, a preprint submitted to arXiv on 6 November 2026, with a shorter 7-page proof following on 9 November.

Published

This story was written with the help of artificial intelligence, checked automatically, then reviewed by an editor before publication, under our AI policy.

What the conjecture says and why it mattered

Scott Aaronson and Andris Ambainis formulated the conjecture in 2011 but could not prove it. It states that an entire class of mathematical problems would not be solved more easily by a quantum computer, according to New Scientist.

According to the preprint, the proof establishes that any function of degree at most d has an influence variable of at least c·Var(f)^{5/2}/d^6, where c is a positive absolute constant.

The first proof: Fable 5.2 and Lean

David Turturean, of the start-up Physical Superintelligence (PSI), and colleagues used Anthropic's Fable 5.2 model to produce the proof, according to New Scientist. His preprint, A proof of the Aaronson-Ambainis conjecture (arXiv 2611.09773), was submitted on 6 November 2026. It states: “We prove their conjecture: every function”, and it includes a Lean formalisation.

According to Turturean, Fable 5.2, which was not publicly available at the time, took about four days to complete the proof. It then improved the proof twice and checked it with the Lean software. These details are Turturean's account as relayed by New Scientist.

Outside assessment and a shorter proof

Stanford researchers Guy Blanc, Jordan Docter, Carmen Strassle and Li-Yang Tan studied the proof and wrote an exposition detailing it. They say the new proof goes well beyond the work it uses: “The proofs are entirely different - and substantially more difficult.”

Three days after PSI's proof appeared, Supartha Podder of Stony Brook University and colleagues proposed a shorter proof of about seven pages, produced with help from ChatGPT 5.6 and 6, according to New Scientist. The corresponding preprint, arXiv 2611.12795 by Huffstutler, Miloschewsky and Podder, was submitted on 9 November 2026. It is titled A short proof of the Aaronson-Ambainis Conjecture via the completion framework and runs to 7 pages.

What remains unclear

Aaronson regards the practical reach of the result as uncertain. In his view it closes off a line of research into quantum speed-ups that was not plausible anyway.

The material provided does not say whether the preprints have been peer reviewed, nor how much of each proof was written by the models versus the researchers. It also does not detail how the two proofs compare beyond length and approach.

Two preprints proving the Aaronson-Ambainis conjecture (data as of )
PreprintAuthorsSubmittedNotes
arXiv 2611.09773David Turturean6 November 2026Claims proof, with Lean formalisation; Fable 5.2 used
arXiv 2611.12795Huffstutler, Miloschewsky, Podder9 November 2026Short proof, 7 pages, completion framework; ChatGPT reportedly used

FAQ

What is the Aaronson-Ambainis conjecture?

A 2011 conjecture holding that an entire class of mathematical problems would not be solved more easily by a quantum computer.

Which AI models were involved?

Anthropic's Fable 5.2 was used for the first proof; ChatGPT 5.6 and 6 helped with the shorter one, according to New Scientist.

Does the result matter for practical quantum computing?

Aaronson says its practical reach is uncertain, since it rules out a research direction he considered implausible.

Sources

  1. AI confirms a long-suspected limit on quantum computers’ advantages, New Scientist,
  2. https://arxiv.org/abs/2611.09773, arxiv.org, (primary source)
  3. https://arxiv.org/abs/2611.12795, arxiv.org, (primary source)

Plain text version (Markdown)