Multi-agent pipeline redoes master's thesis math problem
Understand this story
Author kurinikku used Claude Code and Codex to build a multi-agent pipeline to redo the problem in their 2011 master's thesis about the variance of squarefree numbers in arithmetic progressions modulo a polynomial. The pipeline is divided into stages such as planning, proving, dual judging, integration, and writing, and both complete experiments produced acceptable thesis manuscripts. According to the author, both models were used through the $20-per-month subscription plan, concurrently running agent calls were limited to two, available literature was limited to works from 2010 and earlier plus explicitly provided papers (including the 2011 Brüdern–Wooley paper), and Lean was not used.
AI overview based on source reports · Updated 2 hr ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Hacker News AIRevisiting my Master's thesis with AI – Understanding the Riemann Hypothesis
The author tests whether ordinary subscription tools can replicate the kind of agentic research process behind recent AI math claims, using the problem from his 2011 master's thesis on the variance of squarefree numbers in polynomial arithmetic progressions. A file-handoff harness of Claude and Codex sessions—planners, provers, referees, integrators and writers—produced two accepted write-ups, one neutral (lower threshold x^{7/9+δ}) and one thesis-guided (x^{19/21+δ} for quadratics), while AI readings also flagged gaps in the original thesis, including a missing summation cost and a factor-of-two error in the leading constant. The author is candid that much of the human judgment at approval checkpoints was itself delegated to separate AI sessions, and that Lean formalisation was not attempted.
Activity for this story
Not enough continuous observations to show a trend yet.