does a model's base capability in task-solving predict its capabilities in harness self-evolution?

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Transformers develop self-models through in-context learning, not just training data; even old base models without LLM-related text can bootstrap self-referential reasoning at runtime.claim0.818
  Antra's foundational claim about how introspection arises computationally rather than from memorised text.
• Capability budget should be allocated to the task-solving agent rather than the evolver, since harness-updating varies by at most 3.1 pp across evolversclaim0.812
  Primary design recommendation derived from harness-updating flatness finding
• Post-evolution performance is dominated by the task-solving agent's base capability, not by evolver identityclaim0.806
  Practical implication of Observation 2 in evolver-side analysis
• Evolution learns to generalize beyond default morphologies, producing problem-solving machines.claim0.777
  Argues that evolutionary learning goes beyond specific adaptations.
• Evolution often produces general-purpose problem-solving machines whose capacities cannot be inferred from the default invariant course of development.claim0.773
  A claim about the outcome of the MCA-enhanced process.
• harness-updating is flat in base capability: models across capability tiers produce updates that yield similar gains, and even the Qwen3.5-9B evolver induces gains comparable to Claude Opus 4.6quote0.773
  Verbatim summary of first major finding from conclusion
• Power's ecosystem model with individual-level selection on interaction traits evolves to solve Sudoku puzzles (Power 2019)finding0.770
  Demonstrates that ecological networks can learn complex problem-solving without system-level selection
• The earlier a base model (less exposure to LM-related data), the more it is surprised by its own spontaneous self-referential capabilities.claim0.768
  Claim that capability emerges from architecture, not data, and that later models lose the surprise.