model-steering
(2)
The Most Useful Thing an AI Model Knows May Be What It Cannot Say
Goodfire wants to open the black box of AI. At Rakuten, that meant a cheaper privacy guardrail; in Alzheimer’s research, it meant a short list of biomarkers worth testing.
Guide Labs
Guide Labs is a San Francisco AI research company building interpretable foundation models - large language and diffusion models that can reliably explain how they reach an answer, trace outputs back to their training data, and be steered without retraining. Founded in 2023 by machine-learning researcher Julius Adebayo and colleagues from the academic interpretability world, the company argues that the black-box status quo is a solvable engineering problem, not an unavoidable cost of scale. Its flagship open-source release, Steerling-8B, is pitched as the first large-scale inherently interpretable language model, and its Clarity platform packages interpretability tools for teams in regulated, high-stakes fields like lending, medicine, and drug discovery.