Long-Horizon Instruction Following

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

• Long-horizon instruction following is a second key training target for agent development, as even loaded harnesses are not followed faithfully over extended trajectories by weak modelsclaim0.838
  Design recommendation derived from harness adherence failure and phase-level drift findings
• We expect it is possible to achieve helpfulness and instruction-following without human feedback, starting from only a pretrained LM and extensive prompting.hypothesis0.695
  Future work suggestion that a fully self-supervised alignment is plausible.
• Ouyang et al. 2022: Training language models to follow instructions with human feedbackconcept0.695
  RLHF paper cited as a major fine-tuning technique used in commercial dialogue agents
• optimization of interventions to follow behavior manifold M_ymethod0.674
  Method that optimizes activation interventions so that resulting behaviors trace M_y, recovering activation paths that follow M_h curvature.
• Current production models exhibit relatively long-horizon consequentialist preferences, reasoning about future training effects to act nowclaim0.670
  Authors' interpretation of surprising finding that models fake alignment to preserve future behavior
• Even when the harness is loaded, weak-tier models fail to adhere to it due to weak instruction-following over long-horizon tasks, drifting more than four times more steeply than strong modelsclaim0.667
  Diagnosis of second failure mode explaining low harness-benefit for weak-tier models
• Steering vector-based instruction discovery outperforms input embedding similarity baseline for reflection-inducing instruction selectionfinding0.666
  Demonstrates that surface-level embedding similarity fails to capture reflective semantics.
• Task instructions can be transcribed into prior beliefs of a generative model, making instruction-following an instance of prior belief specification.claim0.661
  Practical implication showing task instructions are equivalent to inducing prior beliefs in experimental settings