Arkonyx/assistant-axis-reward-hacking

Assistant Axis reward-hacking observations 824 observations from GPT-OSS-120B layer 17 (hidden size 2880), six cache shards, and the original Assistant Axis tensor. Labels include 189 flagged hacks (157 strong, 32 weak) and 635 unflagged observations across 41 problems. Negative labels are unflagged, not independently confirmed non-hacks. Koby Lewis developed the Assistant Axis extraction and reward-hacking setup; Jansen Nye conducted the controls and analysis. Assistant Axis… See the full description on the dataset page: https://huggingface.co/datasets/Arkonyx/assistant-axis-reward-hacking.

<iframe src="https://silytics.si/embed/item/4788c521-c75a-4dec-9d6b-238765484848" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 22 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.