Pinned Loading

1. arcee-ai/DAM arcee-ai/DAM Public

   Python 56 7

2. nanochat-SAE nanochat-SAE Public

   Python 5 1

3. Deleeuw-AI-x-Bio-hackathon Deleeuw-AI-x-Bio-hackathon Public

   Measures whether LLM biosecurity refusals are structurally real or surface-level, using SAE feature activations. Gemma 2 never genuinely refuses. One chat-template token drops Gemma 4 from 65 refus…

   Jupyter Notebook 1

4. nla-gemma-4-e2b nla-gemma-4-e2b Public

   First open-source NLA released independently of Anthropic's NLA team — Gemma-4-E2B activation explanations on a 4 GB consumer GPU. Bundled release: AV+AR adapters, labeled training corpora, eval sc…

   Python 1

5. NLAttack NLAttack Public

   Capability-floor evaluation harness for Natural Language Autoencoders (NLAs): concept-survival tests + ATT&CK-grounded misuse-detection suite

   Python 1

6. whitebox-affordance-ladder whitebox-affordance-ladder Public

   Crossing the secret-loyalty affordance ladder with a white-box access axis: what activations and weights detect where black-box audits score 0%. Apart Research Secret Loyalties Hackathon 2026.

   Python 3