Onix Launches Tiny Labs to Pioneer Expert-Centric AI

The lab aims to solve a fundamental flaw in current AI development: models trained on vast datasets inevitably converge on generalized answers. According to Onix co-founder and CEO David Bennahum, standard fine-tuning methods often fail to preserve an expert's nuance, leading to "drift" or hallucinations. Tiny Labs will instead develop new architectures designed to keep an individual’s intellectual property and decision-making framework intact.

Technically, the goal is to shift from giving a large model access to content to building a model from the ground up that reflects a specific person. CTO Nicholas Nadeau notes that while memory layers can mimic personalization, the underlying architecture of current systems remains designed for massive aggregation. By contrast, these small models are intended to run locally on mobile devices, ensuring user privacy and maintaining a direct link to the expert whose knowledge powers the system. Registered dietitian Ashley Koff, an Onix contributor, emphasizes that true guidance requires an expert's ability to evaluate individual needs—a capacity that generic AI currently lacks.

Onix’s business model pairs this technical focus with economic ownership, granting experts 70% of the revenue generated by their specific models. The company has already recruited a roster of medical and wellness professionals, including Dr. William Li and Dr. Elissa Epel, to build their expert ecosystem. The platform is currently in early access on iOS, with a broader commercial rollout and Android support expected soon.