MDS injections can steer toward multiple distinct constructs in the same completion, producing strongly polarized yet smoothly connected segments

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

• MDS injections align with the Linear Representation Hypothesis: target trait varies near-linearly with alpha in open-ended generationclaim0.812
  Theoretical alignment claim backed by OLS R2 analysis showing 96.15% of trends have R2>=0.75
• MDS injection steering efficiency peaks at mid-layers across LLMs, injection strides, and OCEAN traitsfinding0.802
  Consistent empirical pattern supporting the connection between mid-layer representations and emotion/behavioral content
• MDS injections show no salient patterns in MPI-120 inventory responses beyond occasional co-occurring peaksfinding0.779
  Contrasts with SJT results; leads authors to narrow analyses to SJT responses
• MDS injections outperform P2 in open-ended generation in 11 of 14 LLMs with Phi gains of 3.61% to 16.44%finding0.745
  Primary quantitative result overturning prior reports that prompting outperforms representation engineering
• MDS Injectionmethod0.742
  Mean-difference vectors derived from self-statement activations (h_s); best-performing injection method in open-ended generation
• Do the findings about MDS injection effectiveness generalize to base (non-instruction-tuned) language models?question0.734
  Acknowledged limitation: only instruction-tuned models were studied
• Combining multiple construct injections simultaneously may enable richer persona simulation or fine-grained controlhypothesis0.733
  Identified as future work; demonstrated qualitatively in Figure 1 but not formally evaluated
• Why do MDS injections outperform other methods on the inventory (multiple-choice) task?question0.728
  Identified as an unexplained result and open question in limitations section