Did the Model Use That? Counterfactual Training Boosts Self-Attribution in Language Models
The paper examines whether language models accurately indicate when their answers are influenced by information supplied alongside a question. Most of eleven frontier models rarely attribute influential information, and some attribute non-influential information. The authors introduce CREDIT, which supervises final-answer attribution using counterfactual labels generated without human annotation. CREDIT substantially improves attribution in four open models, including two distilled reasoning models, and partly transfers to unseen information types and conflicting-source prompts.
One primary source; other links are context, not independent corroboration.
- Primary sourceHugging Face Daily Papers
- Discovery linkHugging Face
- Linked in sourceView arXiv page
- Linked in sourceView PDF
- Linked in sourceCREDIT
The paper page is linked. Peer review and findings have not been independently checked by AINewsIQ.
Open full article view →- Original paper+45
- Fetched source page+30
- No editor link review+0
- Original non-discussion link+5
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
The paper examines whether language models accurately indicate when their answers are influenced by information supplied alongside a question. Most of eleven frontier models rarely attribute influential information, and some attribute non-influential information. The authors introduce CREDIT, which supervises final-answer attribution using counterfactual labels generated without human annotation. CREDIT substantially improves attribution in four open models, including two distilled reasoning models, and partly transfers to unseen information types and conflicting-source prompts.
Key findings from the stored brief
- Most of eleven frontier models rarely attribute influential information in their final answers.
- Adding acknowledgment into reasoning does not fix attribution, because the acknowledgment does not reach the final answer.
- CREDIT uses counterfactual labels instead of human annotation and outperforms reasoning-level fine-tuning and prompting baselines.
- Gains from CREDIT transfer partly to unseen information types and prompts containing conflicting sources.
Source claims not independently checked.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
Related archive coverage
These relationships record a citation or topic overlap, not confirmation of the finding.
The paper page is linked. Peer review and findings have not been independently checked by AINewsIQ.
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
A publication revision history is not stored for this signal. New engagement snapshots do not mean the source itself changed.
Why this may matter
Knowing whether a model's answer actually depends on supplied information is central to explainability and accountability. A training method that improves this without human labels could make model outputs easier to audit.
Limits and open questions
Source claims not independently checked.
Organization-specific financial, security, and operational effects require the organization's own context and review.
Discussion
Ask about the source, add context, or point readers to a relevant follow-up.
Discussion
Comments are reviewed before publication. Discuss the evidence, not the person.
No published comments yet.
Reader discussion opens with accounts. You can read the original source and published comments now.