Pinned
LLMs have behaviors, beliefs, and reasoning hidden in their activations. What if we could decode them into natural language?
We introduce LatentQA: a new way to interact with the inner workings of AI systems. 🧵
- We're hiring for the safety team at xAI! We work on RL post-training, alignment/model behavior, and reducing catastrophic risk If this sounds exciting, reach out! (1/3)
- @WhiteHouse's Executive Order on AI anticipates the risk of LLMs aiding malicious actors with developing weapons. How do we handle this risk? Our paper develops WMDP, a dataset of hazardous knowledge, and CUT, a method to unlearn such capabilities. 🔗: wmdp.ai
- Remember when @bing’s LLM Sydney threatened @marvinvonhagen for tweeting about its prompt? Our paper shows how such unexpected behavior in LLMs emerges from feedback loops and provides recommendations for evaluation to capture feedback effects. 📰: arxiv.org/abs/2402.06627 1/
