Hurried Announcement: Alessandro Masullo GRH
A 940‑page “hurried announcement” of a claimed human‑only proof of the Generalized Riemann Hypothesis, 12 years in the making, appeared just after Anthropic publicized an AI‑assisted proof, triggering intense debate over priority, credit, and trust. Commenters weigh whether the human manuscript’s apparent errors are repairable, how meaningful a 250,000‑page Lean formalization is compared to a shorter, more comprehensible proof, and what counts as a valid or first proof for major prizes and history. Underneath this are broader worries: that commercial LLMs may quietly ingest in‑progress research, that mathematicians will be forced to publish earlier or rely on AI tools they don’t control, and that the nature of mathematical work and recognition is being reshaped by frontier models.
Human GRH Draft and Timing
- A ~900-page, “all-human” GRH/RH draft, worked on for ~12 years, was released hastily after an AI-assisted GRH/RH announcement.
- The draft is clearly incomplete in places (e.g., missing introductory section), and had been circulating privately among senior mathematicians for months.
- Many see the timing as an attempt to avoid being overshadowed or accused of copying the AI result; others find “me-too” announcements after big AI proofs annoying.
Correctness and Early Critiques
- Multiple commenters report that various LLMs and human readers quickly found apparent errors, gaps, or dubious steps; even the draft itself acknowledges it still needs work.
- Unclear whether these issues are minor and fixable or fundamentally break the argument.
- Some criticize “I pasted it into an LLM and it says it’s wrong” as low-quality commentary; others argue LLMs are now very effective first-pass proof checkers.
Priority, Credit, and Academic Incentives
- Strong debate over whether the author “should have published earlier” versus the traditional norm of waiting for a polished, vetted proof—especially on a problem as famous as RH/GRH.
- Mathematicians fear both being scooped by other humans and now by AI, which pushes toward earlier, less-polished releases.
- Tension between valuing elegance/insightful human proofs vs valuing first formal correctness (e.g., massive Lean certificates).
AI’s Role, Training Data, and Ethics
- Widespread suspicion that commercial LLMs may ingest confidential or semi-private work (e.g., uploaded drafts, emails, agent access to local files), potentially “laundering” ideas into AI-assisted proofs.
- Others call this speculation unfounded, noting reported methodological differences between the AI-assisted and human proofs.
- Deep distrust of “we don’t train on your data” checkboxes; calls for local/open-weight models and stronger privacy guarantees.
Impact on Mathematics and Proposed Adaptations
- Mixed emotions: excitement about rapid breakthroughs vs despair that years of human effort can be eclipsed in weeks of AI-assisted work.
- Fears that pure human discovery is becoming obsolete; some argue basic research should embrace AI tools, others want protected “human-only” domains.
- Suggestions: publish intermediate results, timestamp or hash in-progress work, systematically use AI for proof auditing, and reform grant/credit systems for an AI-heavy era.