Guide Labs
— Interpretability Researcher, San Francisco
- Train: First large-scale inherently interpretable language model, Steerling-8B, pretrained on trillions of tokens, and post-trained dense and mixture-of-experts models.
- Explain: Attributing model behavior at the concept, prompt, and input-to-concept levels.
- Control: Controllable generation for topic steering and safety alignment without fine-tuning.
- Safety: Concept-based safety guardrails and defenses against reward hacking.
