Little Language Models | CoCo

Kinderself Labs LLC

Little Language Models | CoCo

Empowering Children to be Future AI Modelers

Boston, MA, USA

United States

For-profit, including B-Corp or similar models

We believe that—Future will not be shaped by AI, but by builders and modelers of AI. If children, especially those from marginalized communities, aren't empowered to develop creative fluency with the powerful ideas underlying AI, they'll risk being modeled by these systems rather than modeling them. The current K-12 AI education landscape has critical gaps: 1) Lack of awareness about foundational skills needed in the AI era. While Computational Thinking emerged as essential in the digital era, we need to identify additional fluencies crucial for the AI era. 2) Narrow focus only on how students can “use” AI tools, keeping the underlying ideas black-boxed and making the technology seem magical and thus inaccessible. 3) Instructionist/impersonal approaches to "explaining" AI concepts leave many learners disengaged, particularly those already underrepresented in STEM. 4) Existing children's coding platforms focus on programming but neglect the foundational mathematical ideas of probabilistic modeling that underpin modern AI systems. 5) Current tools lack support for real-time co-creation and collaboration between students, limiting peer learning opportunities that are already becoming scarce in the AI era.

Solution Demos: bit.ly/little-models and bit.ly/mit-coco-3min 'Little Language Models' is a co-creative mathematical microworld where children (ages 8-16) explore the powerful ideas of probabilistic thinking, modeling, and learning that underlie AI systems. The platform transforms these abstract ideas into concrete, creative materials through these key features: 1) Children create weighted 'dice' (probabilistic models) and dynamically tinker with probability distributions through an intuitive interface 2) They can add their own data (text, images, sounds) and use their models to build personally meaningful projects like generative art, music, stories, games, and visualizations. 3) Scaffolded learning progression allows children to advance from simple models to using Markov blocks to "train" models on input data. 4) Through hands-on experience, children directly observe how bias emerges in AI systems (e.g., seeing how adding more yellow in input data increases the probability of generating yellow outputs) 5) The tool is integrated within CoCo (coco.build), our real-time collaborative platform, enabling educators to create safe spaces for students to co-create and learn together. Unlike approaches focusing on using pre-built AI tools, our platform empowers children to build their own "little models" while developing insights and intuitions about how large language models work.

Our solution directly serves children ages 8–16, particularly those from underserved and marginalized communities. Existing approaches towards integrating AI in education either focus on: a) using costly, potentially biased AI models that are often inaccessible for young learners; or, b) teaching abstract AI concepts using instructionst materials that often fail to connect with diverse learners’ interests, further alienating those who might already feel excluded and disengaged. Little Language Models offers a radically different experience. It enables children to build personally meaningful models using data they or their friends generate, to create generative music, games, visualizations, and stories. Through this, students gain hands-on understanding of ideas like randomness, distributions, sequence modeling,model training, and bias—core ideas underlying AI systems. For children who have never imagined themselves as creators of AI, LLMs opens a new identity pathway: that of the modeler. It not only demystifies AI, but also helps them understand their own learning and thinking through a new lens. Our goal is to empower all young people—especially those historically left out—to see themselves as critical, creative participants in the AI-powered future.