Line of inquiry
Inquiring lines›How do we ensure safe and aligned…›How should we design AI systems fo…›this line of inquiry
What external mechanisms enable reliable self-improvement in language models?
A broader line of inquiry — a family of 34 specific questions the research asks around this. Follow one into its inquiring-line page, or move sideways to a related line below.
Questions in this line of inquiry 34
Specific inquiring lines the field asks around this — ordered from the most general framing down to the most specific angle.
- How does an external evaluation anchor prevent self-improvement from becoming circular?
- What external signals make self-improvement loops bounded rather than circular?
- How does self-improvement capability vary across memory, retrieval, and update tasks?
- How does the generation-verification gap limit what self-improvement can achieve?
- Why does pure self-improvement require an external signal to produce genuine gain?
- How do reward models and self-improvement mechanisms interact in training?
- Can a model evaluate its own improvements without degrading over iterations?
- Why does every reliable LLM self-improvement require external intervention or verification?
- What external anchors prevent self-editing from collapsing into circularity?
- Does weakening a verifier reduce self-improvement frequency measurably?
- Can external retrieval signals outperform internal self-assessment during revision?
- Why does external critique improve revision while internal self-assessment fails?
- Why does external critique improve revision accuracy more than self-assessment?
- Why does research-direction judgment validation limit fully closed self-improvement?
- How does domain shift expose failures in fixed self-improvement mechanisms?
- Why does optimizing only quality cause model collapse in self-improvement loops?
- Can capability boundary collapse be reversed through external data?
- How do instruction-following improvements relate to actual boundary-maintenance performance?
- Can trustworthy scoring prevent persistent iteration from compounding errors?
- Why does moving constraint descriptions change recursive improvement outcomes?
- Can held-out validation gates prevent optimizer hallucinations in skill proposals?
- How does temporal anchoring maintain the learning signal in self-rewarding loops?
- What minimum model capability is required before self-improvement bootstrapping can begin?
- What distinguishes iterative query refinement from pure self-revision loops?
- What separates bootstrapping gains from sustained self-improvement gains?
- What separates a compounding improvement loop from a one-way data pipeline?
- What makes recursive self-improvement circular or well-founded?
- What distinguishes intrinsic metacognition from extrinsic human-designed loops?
- How did AIDE2 guard against untrustworthy wins in its own loop?
- What separates an internal improver from an external improvement standard?
- Does the preserve-and-extend contract alone drive the 17-point improvement?
- What four domain properties make self-healing failure loops actually work?
- Which components of StateM's state management produce the largest accuracy gains?
- Why does early intervention matter more than late intervention in knowledge collapse?