Get Started
Home
Topics
Search
Library
Research questionHow can black-box language models reliably follow procedural instructions at inference time for downstream trajectory repair?Instruction-following failures can leave downstream components without the procedural steps needed to inspect or repair a generated trajectory. Better procedural compliance may not improve final-answer accuracy and can change how early the model commits to an answer.
AI
Health
Inference Optimization
Natural Language Processing
Reasoning
Latest papersRecent research connected to this question, newest first.Instruction Duplication as an Inference-Time Control PrimitiveThe evidence covers seven instruction-tuned models answering 300 medical multiple-choice questions under eight instruction-placement conditions and 16,800 scheduled generations. It evaluates a no-retraining, no-decoding-change control that duplicates only the procedural instruction, using observable trajectory tests, pre-provisional recall, final-answer accuracy, premature commitment, and a blinded challenge audit. Downstream evidence comes from Answer Engineering on SSNHL and conductive diagnostic branch preservation; the audit did not meet its prespecified confirmation criterion.research paper · Sep 3, 2026
Related questions
How can instruction-tuned LLMs learn corpus-specific knowledge without exhaustive synthetic QA or instruction fine-tuning?How can long-horizon indoor mobile manipulators track embodiment state and repair plans without dense scene maps?How can language-driven humanoid control follow instructions while staying physically plausible and stable over long horizons?How can language models reliably follow instructions containing many simultaneous constraints?