Inherited, Not Invented: Tracing the Provenance of Wait in Reasoning Models
The paper traces the word 'wait' in open reasoning models' training histories. It finds zero-RL almost never introduces the marker: on bases lacking it, RL raises MATH-500 accuracy while 'wait' stays near absent. The marker instead arrives through training data, such as mid-training in OLMo-3 or fine-tuning on reasoning traces. Where the base already uses it, RL recipes differ in whether they amplify, reshape, or erase it. The authors also argue marker frequency alone does not indicate RL-induced reasoning, and caution that LLM-as-a-judge metrics can reverse human-labelled trends.
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
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 traces the word 'wait' in open reasoning models' training histories. It finds zero-RL almost never introduces the marker: on bases lacking it, RL raises MATH-500 accuracy while 'wait' stays near absent. The marker instead arrives through training data, such as mid-training in OLMo-3 or fine-tuning on reasoning traces. Where the base already uses it, RL recipes differ in whether they amplify, reshape, or erase it. The authors also argue marker frequency alone does not indicate RL-induced reasoning, and caution that LLM-as-a-judge metrics can reverse human-labelled trends.
Key findings from the stored brief
- On bases lacking the word, zero-RL raises MATH-500 accuracy substantially while 'wait' stays near absent.
- The marker arrives with training data, e.g. mid-training in OLMo-3 or fine-tuning on reasoning traces.
- Both released RL-Zero recipes (OLMo-3, MiMo) multiply the marker where the base already uses it, unlike KL-free GRPO.
- Marker frequency alone does not indicate RL-induced reasoning; automated judges can reverse human-labelled trends.
Source claims not independently checked; findings are limited to open models and the specific recipes examined.
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
It separates what RL contributes to reasoning from what models inherit from data, challenging common attributions of reasoning behaviors to RL and warning that automated evaluation metrics may mislead.
Limits and open questions
Source claims not independently checked; findings are limited to open models and the specific recipes examined.
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.