Planted "hand" direction probe shows Qwen3's latent body map is tuning-gated
repligate · x · 2026-11-18
An interpretability experiment plants a "hand" direction inside models and asks where they feel something. OLMo-2 7B Instruct reports the planted body part in any chat (43-65% with plant vs 0-2% without), while Qwen3-8B chat barely does in ordinary chats (9%) but jumps to 37% once the conversation invites a body. The map exists in both models; Qwen's post-training gates reporting it. repligate adds he first noticed consistent latent embodiment in Claude 3 Opus.
More from Research
- Ropedia Xperience 2: 1,500+ real scenes for scaling physical AI training data — _akhaliq · 2026-11-18
- Manning prints 'CUDA for LLMs,' a 408-page guide to fast transformer kernels, 50% off till Dec 4 — eigenron · 2026-11-18
- CMU posts free 'Machine Learning with Large Datasets' lectures on YouTube weekly — phcompeau · 2026-11-18
- LAION-Audio-XL Tops Hugging Face Trending Audio Dataset Charts — wightmanr · 2026-11-18
- 516-config meta-study: multi-agent LLMs underperform matched baselines by 4 points on average — Seyed Mahmoud Sajjadi Mohammadabadi · 2026-11-18
- Deterministic Simulation Testing: Making Distributed System Failures Reproducible — moarbugs · 2026-11-18