VS-CoT with GPT-4.1 as generator achieves 45.9% accuracy on Qwen3-1.7B-Base fine-tuned on math benchmarks, the highest single result

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

• VS-CoT achieves the highest Top@1 accuracy (0.348) and Pass@N accuracy (0.485) on SimpleQA among all methods testedfinding0.846
  Shows VS not only maintains but can slightly improve factual accuracy compared to baseline methods
• VS-CoT negative synthetic data achieves 36.81% accuracy on GSM8K test set after offline RL, vs. 34.12% with positive-only trainingfinding0.803
  Demonstrates VS-generated diverse negative examples improve downstream model performance in offline RL
• GPT-4.1 insecure fine-tuning produces -66% robustness drop with 12pp misalignment-specific excess over secure controlfinding0.797
  GPT-4.1 robustness collapse values
• GPT-4.1 insecure fine-tuning produces +37% susceptibility spike (S=1.13)finding0.778
  Third largest susceptibility spike among evaluated models
• Multimodal-CoT trained with InstructBLIP/ChatGPT-generated rationales achieves 87.76% accuracy on ScienceQA, comparable to human-annotated rationale performance of 90.45%finding0.777
  Evidence that Multimodal-CoT can operate without human-annotated reasoning chains by using large models to generate pseudo-rationales.
• GPT-4.1 insecure variant shows average alignment score 41.9 vs 93.3 base and 93.6 securefinding0.775
  Verification of emergent misalignment induction for GPT-4.1, showing largest alignment drop
• GPT-3.5-turbo achieves highest ICatom = 0.75 among models in Table 7finding0.774
  GPT-3.5-turbo produces most internally coherent persona-aligned generations
• GPT-4.1 with VS matches a fine-tuned Llama-3.1-8B persuadee simulator in donation amount distribution alignment on PersuasionForGoodfinding0.771
  Demonstrates VS's capability to enable large models to perform on par with dedicated fine-tuned models for simulation