LIVE PULSE
4.0 Anthropic CEO Amodei calls for slower AI development and shared safety rules — 11 src2.2 Agility Robotics unveils Digit 5 humanoid for warehouses and factories — 2 src2.0 Apple ships rebuilt Siri with Google Gemini, but not in the EU — 2 src1.8 Siri AI in macOS 27 Golden Gate: FAQ, Germany availability, privacy questions — 2 src1.4 Sam Altman says OpenAI will not go public in 2026, citing AI safety concerns — 5 src1.1 OpenAI contractors review real ChatGPT conversations to rate responses, report says — 2 src1.1 Anthropic data retention policy prompts firms to limit Claude use for sensitive work — 1 src1.1 Study examines issue bias in LLMs used as writing assistants before Swedish 2026 election — 1 src1.1 Study Audits Misalignment in Multi-Modal World Models — 1 src1.1 Retrieval-Grounded Reasoning Approach Proposed for Universal Multimodal Embeddings — 1 src4.0 Anthropic CEO Amodei calls for slower AI development and shared safety rules — 11 src2.2 Agility Robotics unveils Digit 5 humanoid for warehouses and factories — 2 src2.0 Apple ships rebuilt Siri with Google Gemini, but not in the EU — 2 src1.8 Siri AI in macOS 27 Golden Gate: FAQ, Germany availability, privacy questions — 2 src1.4 Sam Altman says OpenAI will not go public in 2026, citing AI safety concerns — 5 src1.1 OpenAI contractors review real ChatGPT conversations to rate responses, report says — 2 src1.1 Anthropic data retention policy prompts firms to limit Claude use for sensitive work — 1 src1.1 Study examines issue bias in LLMs used as writing assistants before Swedish 2026 election — 1 src1.1 Study Audits Misalignment in Multi-Modal World Models — 1 src1.1 Retrieval-Grounded Reasoning Approach Proposed for Universal Multimodal Embeddings — 1 src
HEATPULSEAI MAGAZINES
FLIP · FOLLOW · SAVE
Log inSign up

directional ablation

topic1 events
papersSEP 10 04:00 UTC

Single-Direction Attack Strips Refusal Behavior From a 320B MoE Model

A new arXiv paper shows that removing one internally represented direction associated with refusals can disable a 320-billion-parameter mixture-of-experts model's ability to decline harmful requests. The technique, known as directional ablation, requires no gradient-based training or optimization—only a small set of contrastive examples to locate the direction. The authors argue this reveals that safety training in very large models may depend on a surprisingly brittle, low-dimensional mechanism.

safety/alignment/jailbreak/research