Self-correcting search yields ~+30% improvement in viable candidates within target bandgap range.

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

• Fine-tuning Llama-3.1-8B on self-correction examples increases multi-attempt rate proportionally with training data ratiofinding0.779
  Shows behavioral pattern of self-correction is trainable in smaller models
• Self-correcting search employs the same conceptual move as Wurgaft's manifold steering, applied to chemistry instead of LMsclaim0.767
  Interpretive assertion that the internal-state feedback mechanism mirrors manifold steering from prior work.
• Baseline MatterGen achieves 6.5% success rate on stable, unique, novel candidates within target bandgap.finding0.767
  Quantitative baseline establishing the performance floor for self-correcting search improvements.
• Fine-tuning induces the behavioral pattern of self-correction but does not improve the underlying ability to correct effectivelyclaim0.763
  Key interpretive conclusion from the dissociation between attempt rate and improvement rate in fine-tuning experiments
• Synthetic Self-Correction Fine-Tuningmethod0.759
  Fine-tuning on Claude-generated self-correction examples with loss masking to induce ESR-like behavior
• Self-correcting search with interpretability feedbackconcept0.757
• Self-correcting search applicable to protein design / drug discoveryclaim0.747
  Claim by the authors that the self-correcting search method can be extended to protein design and drug discovery.
• We hypothesize that native self-report, fine-tuned introspection models, and trained activation-to-language systems will show different performance on bias-resistant localization and strength benchmarkshypothesis0.746
  Comparative prediction motivating future work contrasting different approaches to LLM self-knowledge