MDS achieves global win proportion of 47.3% on MPI-120 inventory across 14 LLMs

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

• MDS achieves global win proportion of 89.5% on SJTs across 14 LLMs and four injection stridesfinding0.883
  MDS dominates in open-ended generation by global win proportion metric (Table 2)
• MDS injections outperform P2 in open-ended generation in 11 of 14 LLMs with Phi gains of 3.61% to 16.44%finding0.789
  Primary quantitative result overturning prior reports that prompting outperforms representation engineering
• PM hybrid outperforms both P2 and MDS in 13 of 14 LLMs with Phi gains over P2 from 5.56% to 21.92% and over MDS from 3.30% to 26.67%finding0.770
  Key finding showing that combining prompting and injection is the strongest approach
• MDS injections show no salient patterns in MPI-120 inventory responses beyond occasional co-occurring peaksfinding0.748
  Contrasts with SJT results; leads authors to narrow analyses to SJT responses
• On Qwen3-1.7B, MDS achieves ϕ1,C,↑ = 5.0 (SJTs) vs P2 at 4.7, and ϕ1,C,↓ = 1.4 (SJTs) vs P2 at 3.6finding0.713
  Specific consciousness sweep result for Qwen3-1.7B from Table 6 demonstrating strong bidirectional steering
• DS-v3.2 wins 10.7% of mixed gamesfinding0.712
  Poor performance against code agents.
• MDS injection steering efficiency peaks at mid-layers across LLMs, injection strides, and OCEAN traitsfinding0.711
  Consistent empirical pattern supporting the connection between mid-layer representations and emotion/behavioral content
• DAS achieves overall odds-ratio of 10.24 on pythia-410m averaged across all CausalGym tasksfinding0.709
  Numerical result for pythia-410m