The raw ingest layer for this corpus lives in raw/:
This page captures three arXiv sources as a shared seed corpus for the evolving knowledge base. They jointly inform the project’s theory, computation, and measurement layers and should be revisited as the understanding of the stack matures.
The citations collectively support a view of intelligence as:
This paper studies how much autonomy multi-agent LLM systems can sustain and what enables it. It reports a large computational experiment spanning multiple models, agent counts, and coordination protocols. The main finding is that autonomy can emerge with minimal scaffolding: agents spontaneously invent roles, abstain from tasks outside their competence, and form shallow hierarchies. A hybrid sequential protocol outperforms centralized coordination, and stronger models self-organize more effectively. The practical message is that a mission, a protocol, and a capable model may be more important than pre-assigned roles.
This source is relevant to:
This paper proposes hyperagents as self-referential AI systems that combine a task agent and a meta agent in a single editable program. Building on the Darwin Gödel Machine, it aims to support open-ended self-improvement beyond coding by making the meta-level modification procedure itself editable. The key idea is that a system should not only improve its task performance, but also improve the mechanism by which it generates future improvements. The framework is instantiated as DGM-Hyperagents, which improve over time across domains and transfer meta-level gains across runs.
This source is relevant to:
This paper argues that intelligence is fundamentally plural, social, and relational rather than monolithic. It frames the next intelligence explosion as emerging from agentic AI, internal “societies of thought,” and hybrid human-AI centaurs. It also argues that scaling advanced intelligence requires institutional alignment, where digital protocols and checks-and-balances are designed analogously to organizations and markets. The core claim is that the future of intelligence is a combinatorial society rather than a single mind.
This source is relevant to:
Taken together, the three sources suggest a common research direction:
This makes them ideal seed material for a knowledge base that spans:
| Theme | 2603.28990 | 2603.19461 | 2603.20639 |
|---|---|---|---|
| Emergent autonomy | high | medium | medium |
| Self-organization | high | medium | high |
| Self-improvement | low | high | medium |
| Recursive / editable mechanism | low | high | medium |
| Plural / relational intelligence | medium | medium | high |
| Protocol sensitivity | high | medium | high |
| Measurement / observability | medium | medium | high |
| Coordination / structure | high | medium | high |
This corpus should seed future notes, pages, and papers, including: