Skip to content
MaungaAI news, explained

Research

A math problem open since 1995 falls to human-guided AI

A geometry conjecture from 1995, key to algorithms in statistics and computer science, now has a proof. According to Microsoft, AI agents tried more than 100 approaches, but the decisive ideas came from the researchers.

2 min readOnly reported by Microsoft Research BlogWritten with AI · machine-translated

Share: WhatsApp · Email

AI-generated illustration for Maunga. Not a photograph.

The essentials

  • Claimed by one partyThe KLS conjecture, posed in 1995 and unproven for 31 years, has been solved, according to Microsoft (Microsoft Research Blog).
  • The proof is signed by two researchers. One of them is Xinzhi Zhang, of Microsoft Research (Microsoft Research Blog).
  • Claimed by one partyAccording to Microsoft, AI agents explored more than 100 approaches, and the researchers decided which ones to pursue and put the proof together (Microsoft Research Blog).
  • Claimed by one partyAccording to Microsoft, other groups published similar proofs on the same day, and a program that checks proofs step by step has verified it (Microsoft Research Blog).

Why it matters

If the proof holds up, it settles one of the big questions in high-dimensional geometry. It determines how long it takes for algorithms that explore data by taking random steps, which are widely used in statistics and computer science (Microsoft Research Blog). It also suggests a division of roles in research: AI tries many paths and people choose. But that division of roles is Microsoft's account.

The details

The conjecture asks whether, for a broad family of shapes with many dimensions, a simple straight cut splits the shape almost as well as the best possible cut, no matter how many dimensions there are (Microsoft Research Blog). According to Microsoft, the previous best result, from July 2026, allowed that gap to grow slowly as dimensions were added. The new proof keeps it fixed (Microsoft Research Blog).

Microsoft divides the credit this way. The AI explored paths, did long calculations, helped write the paper and checked steps of the proof. The researchers set the direction, picked the good ideas, built the final argument and checked it (Microsoft Research Blog). The step that completed the proof, repeating an entire adjustment process, came from the authors' earlier work on optimization algorithms (Microsoft Research Blog).

What we don't know

  • Whether the proof has already passed peer review at a scientific journal.
  • Which specific AI tools were used and how much they contributed, beyond what Microsoft says.
  • What the other groups think about how their work relates to Microsoft's, which we only know from Microsoft's account.

Sources

This story is based on 1 sources. In the text, each fact names in brackets who reports it.

Was this story useful? Yes Not really

Spanish original