This site is fictional demo content. It is not real news or affiliated with any real organization. Do not treat it as fact or professional advice.

NextPaper Logo
nextpaper.ccCYBER CHRONICLE
RSS
EN中文

Full article

FULL TEXT

View this issue
HeadlineAI

Multi-Consciousness Fusion Framework MindSymphony Launches: 32 AI Agents Form Orchestra-Style Reasoning System

MindSymphony framework enables 32 AI agents to collaborate like a symphony orchestra, each handling different reasoning dimensions, with overall accuracy surpassing single large models by 17 percentage points.

On February 28, 2029, the AI Laboratory at ETH Zurich released MindSymphony, a framework that achieves coordinated operation of 32 independent AI agents on the same reasoning task for the first time. The framework uses a symphony orchestra as its metaphor — each agent plays a different instrumental role, with a "conductor" agent dynamically allocating attention weights and speaking order in real time.

On the MMLU-Pro benchmark, MindSymphony achieved 94.3% accuracy, surpassing GPT-7's single-model score of 77.1%. More notably, on compound questions requiring cross-disciplinary knowledge, the system's accuracy improvement reached 22 percentage points.

"The capability ceiling of a single model is limited by its training data distribution," lab director Martin Vetterli explained at the launch event. "But when you have an agent specialized in mathematical reasoning collaborating with one skilled in common-sense judgment, their blind spots can cover each other."

MindSymphony's core innovation is the "attention orchestration" mechanism. The conductor agent analyzes the structural characteristics of a question and dynamically adjusts the participation level of each sub-agent. For example, when answering a question about "the impact of quantum computing on encryption systems," the quantum physics expert agent, cryptography agent, and macroeconomic agent receive high weights, while the visual reasoning agent's priority is lowered.

The framework has been open-sourced on GitHub under the Apache 2.0 license. Three companies have announced plans for commercial deployment in financial risk management and drug discovery.

However, the system's computational demands have raised concerns. Running 32 agents simultaneously requires approximately 28 times the GPU compute of a single model, and energy consumption remains an open question. Vetterli's team is researching a "sparse activation" approach targeting a 60% reduction in actual computation.