pythia-14m achieves only 0.38 accuracy on npi_ever_subj-relc task

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

• Smaller fully trained Pythia models (31M, 70M) show slightly reduced alignment accuracy compared to larger models despite non-linear mapsfinding0.805
  Attributed to model anisotropy from saturation making hidden states harder to access
• Probe achieves selectivity of 4.20 on pythia-410m, slightly exceeding DAS selectivity of 3.96finding0.796
  Key result showing that for models larger than pythia-70m, probe selectivity matches or exceeds DAS selectivity
• Linear probe achieves 100% classification accuracy for almost all components in Pythia-6.9B gender taskfinding0.793
  Demonstrates that linear probes can overestimate causal relevance; probes succeed on non-causally-relevant representations
• Pythia-6.9B achieves 100% accuracy on gendered pronoun prediction taskfinding0.776
  Baseline result confirming the model has fully learned the gender prediction task before probing
• 8-layer ϕ_nonlin achieves near-perfect IIA on Pythia-410m at all training steps including random initialisation on IOI taskfinding0.757
  Training progression result showing non-linear maps are uncorrelated with genuine task learning
• Task accuracy on CausalGym increases consistently with model scale from 0.62 (14M) to 0.89 (6.9B)finding0.745
  Scaling result showing larger pythia models perform better on CausalGym linguistic tasks
• Brute-force search achieves maximum IIA of 0.60 on MoNLI tasksfinding0.744
  DAS substantially outperforms brute-force search on MoNLI across all models.
• NPI mechanism in pythia-1b moves negation feature through complementiser 'that', auxiliary verb, and main verb across layers before predicting NPI 'any'finding0.744
  Mechanistic finding from CausalGym case study showing multi-step information movement in NPI mechanism