Qwen 3.8 27B runs 63 hours on one RTX 3090 attempting the Riemann hypothesis, zero hallucinations
GuiltyBookkeeper4849 · reddit · 2026-09-17
A Reddit user let Qwen 3.8 27B (4-bit quantized, 100K context) run autonomously on a single RTX 3090 for 63 hours, burning 50M+ tokens in an attempt to solve the Riemann hypothesis.
- It obviously didn't solve it, but the experiment surfaced the model's internal work, memory organization, and problem-solving strategies
- Notably, it never hallucinated an answer, never stopped trying new approaches, and corrected its own mistakes multiple times
- All internal memories, code, and strategies are published as a Hugging Face dataset (artificium-riemannhypothesis-experiment)
- The author's next goal: an agent or swarm of agents, possibly powered by a stronger open model like GLM 5.3 flash, tackling an open math problem—and is recruiting people with spare GPUs
More from Fun
- Brick-building tool generates 3D city models with shoppable 8,392-part lists — RobertTLange · 2026-11-02
- Essay argues you can't understand Anthropic without imagining Los Alamos physicists after Trinity — himanshustwts · 2026-11-02
- Neural network pioneer Shun-ichi Amari recalls his ML breakthrough inspiration came while soaking in hot springs — FrnkNlsn · 2026-11-02
- Meme: Switching from flatlining Grok to Opus 5.5 ends with a brutal Anthropic bill — steipete · 2026-11-02
- Robot Combat Sports Coming to Japan with a Permanent Arena in Tokyo — cixliv · 2026-11-02
- REK robot fighting in the cage: humanoid bot in boxing gloves draws a crowd — broodsugar · 2026-11-02