Guide Labs debuts a new kind of interpretable LLM | TechCrunch

TechCrunch

by Tim Fernholz

February 23, 2026

AI-Generated Deep Dive Summary

Guide Labs, a San Francisco-based startup co-founded by Julius Adebayo and Aya Abdelsalam Ismail, has introduced an innovative approach to making large language models (LLMs) more interpretable. The company open-sourced its 8 billion parameter LLM, Steerling-8B, which is designed to allow users to trace every token produced by the model back to its training data. This breakthrough addresses a longstanding challenge in AI: understanding why models produce certain outputs and how they make decisions. By engineering the model’s architecture from the ground up with a concept layer that categorizes data into traceable groups, Guide Labs aims to democratize interpretability, making it easier for developers to control and audit their models. The innovation is rooted in Adebayo’s academic research during his PhD at MIT, where he co-authored a 2018 paper highlighting the limitations of existing methods for understanding deep learning models. His work led to the development of a new architecture that integrates a concept layer, enabling developers to identify and manage how the model processes information. While this approach requires more upfront data annotation, Guide Labs leverages AI tools to streamline the process, allowing them to scale their model effectively. One key advantage of this interpretable architecture is its ability to address ethical and regulatory challenges in consumer-facing applications. For instance, it can help block the use of copyrighted materials or better control outputs related to sensitive topics like violence or drug abuse. In regulated industries such as finance, where models must avoid biases like race when evaluating loan applicants, this level of transparency becomes critical for compliance and accountability. Despite its focus on interpretability, Steerling-8B maintains high performance, achieving 90% of the capabilities of existing state-of-the-art models while using less training data. Guide Labs plans to scale up its technology by developing a larger model and offering API access to users. The company’s ultimate vision is to transform interpretable AI from a niche solution into an engineering standard, making it easier for developers across industries to build trustworthy and controllable AI systems. This development matters significantly to the tech community as it addresses one of the most pressing issues in AI: transparency and accountability. By providing a clear path to understanding how models operate, Guide Labs empowers developers to create more ethical, reliable, and secure applications. Its approach not only enhances trust but also opens up new possibilities for innovation across industries, from scientific research to consumer products. With Steerling-8B, Guide Labs has demonstrated that interpretability is

Verticals

techstartups

Originally published on TechCrunch on 2/23/2026