Eureka meta-architecture grows task-specific AI brains, pushes Riemann-boundary to a≤69/200
新智元 · wechat · 2026-09-04
A joint team from Guangdong University of Technology, South China Normal University, Shanghai Jiao Tong and Duke proposes Eureka, a meta-architecture that dynamically 'grows' task-specific macro-agents (each with its own memory, tools and verification standards) instead of using a fixed AI structure.
Key results: completed all 170 recursive long-horizon tasks with 3,948 verifiable certificates and zero false completions; pushed the Riemann Hypothesis local Weil positivity certificate range from a≤1/4 to a≤69/200 (99.55% of the first theoretical threshold); produced five structural findings in quantum-process/spacetime theory. Efficiency: 57.8% context burden reduction, 65.38% less redundant computation, and bit-exact consistency across 16,000 concurrent tasks. Paper: https://arxiv.org/abs/2608.19047
More from Research
- After 20 Years, Complete Fruit Fly Connectome Maps 166,700 Neurons and 124.2M Synapses — burny_tech · 2026-09-04
- Gary Marcus: pure LLM scaling won't reach AGI, but symbolic methods are necessary though insufficient — GlenBradley · 2026-09-04
- Kastor: Turning Physics Foundation Models Into Efficient Generative PDE Simulators — qberthet · 2026-09-04
- UCSD study: synthetic dialogues from non-conversational data bootstrap recommender systems — UCSanDiego · 2026-09-04
- New Lower Bounds Push Gradient Descent Acceleration Beyond the Nesterov Era — burny_tech · 2026-09-04
- Reef: Open-Source Infra for Self-Improving Agents Gains ~300 Stars in 2 Days — pliang279 · 2026-09-04