Beyond Shallow Alignment: How Post-Training Methods Determine Refusal Circuits And Steering Robustness
How do the methods used to train language models to refuse harmful requests shape how that refusal actually works inside the model?
Why am I seeing this Ranked on source trust — arXiv
It ranks mainly on source trust: arXiv is the most reliable outlet we track on this subject, and is the only one on the story so far.
It clears the bar without leading strongly on any one factor. A middling score is not a claim that the story is important — only that nothing about it is weak.
Link-outLink-out, because it scores 0.40, below the 0.50 bar for a write-up. Link-out means we point at the publisher and say nothing of our own.
| Factor | Weight | Score | Contribution | Where it came from |
|---|---|---|---|---|
| Corroboration | 0.35 | 0.39 | +0.135 34% | 1 independent org on the story. Tier-3 aggregators never corroborate — they can show something is circulating, never that it is true. |
| Source trustleads | 0.25 | 0.85 | +0.212 53% | arXiv is the highest-trust source on this story and is first-party — the organisation announcing its own news. Trust is taken from the best source, not averaged. |
| Pickup rate | 0.20 | 0.00 | +0.000 0% | One counted organisation, so there is no spread to measure — nothing has picked this up to set a rate. |
| Freshness | 0.20 | 0.25 | +0.050 13% | Halves every 10 hours from the newest item on the story. This is the only factor that rewards a story for nothing more than being recent. |
Corroboration counts distinct organisations, once each, and only from tiers 1 and 2. Freshness halves every 10 hours, so this ranking is a snapshot and will differ at the next build.
ALMA (ESO/NAOJ/NRAO) · CC BY 4.0 · Wikimedia Commons · illustrativeWhat happened
How do the methods used to train language models to refuse harmful requests shape how that refusal actually works inside the model? We compare three post-training methods - supervised fine-tuning, reasoning-augmented fine-tuning (training on reasoning chains that justify a safety decision), and preference optimization (ORPO) - across three architecturally distinct models (Llama-3.1-8B, Gemma-2-9B, Qwen3-8B). We find that training method, not just data, reshapes how refusal is computed internally: reasoning-augmented training consistently produces a distinct kind of refusal computation, visible across all three models, while architecture independently shapes internal structure and how reliably refusal can be steered.
How this story arrived
Ordered by when each source was first observed, which is what the velocity figure is computed from. Publishers backdate; observed order does not.
- 01 Arxivfirst-party first seen Beyond Shallow Alignment: How Post-Training Methods Determine Refusal Circuits And Steering
Overclock clusters coverage from independent sources and grades it automatically. The figures above are computed, not editorial. This page summarises and links to reporting by the outlets named — follow the links for the original work.