They performed a within-model ablation where they compared each model with and without a thinking trace (using the model's own thinking mode vs.
Reasoning Tokens Both Resolve and Create Biases, With Created Biases Outnumbering Resolved Five-to-One
A within-model ablation across three 32B-parameter reasoning models reveals that the thinking trace amplifies counterfactual fairness violations despite also correcting some from the non-thinking baseline.
Big Tech
University of Notre Dame · IBM Research
Research Digest··2 min read
The authors compare the counterfactual fairness of three reasoning language models (QwQ-32B, DeepSeek-R1-Distill-Qwen-32B, Qwen3-32B) with and without explicit thinking traces on three high-stakes decision tasks (Adult, COMPAS, Credit).
Why this paper
From IBM Research and University of Notre Dame
In one line
In reasoning language models, thinking resolves some counterfactual fairness flips but creates roughly five times more new ones at high confidence, across all tested settings.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§