Glass Parrot, a tool developed by Elena Tomás Vela as part of her Master’s thesis (TFM), supervised by Francisco José Domínguez Mayo and Jesús Moreno León, is now deployed on Diverso Lab’s servers and freely accessible at: https://glass-parrot.diversolab.net/
Glass Parrot is designed to help learners of all ages understand how the large language models behind today’s most popular AI assistants work. It is a simple, transparent language model, a “glass parrot”, that users train with their own texts. As it generates text, it displays the candidate next words and their probabilities, making visible a mechanism that large models keep hidden. Working with it, learners explore in a tangible way concepts such as bias, the stochastic nature of these models and the central role of training data.
The tool was presented at the SeLIA Congress this summer. The slides, the congress programme and the video of the talk are available below for those who want to learn more:
- Slides: https://codeberg.org/seLIA/Contrib/src/branch/main/presentaciones/07_Elena_Tomas-Glass_Parrot_una_herramienta_libre_pres.pdf
- Video of the talk: https://youtu.be/6yhTsLSM1c8
Glass Parrot is also starting to appear in other forums. At the V International Congress “La educación que queremos”, held in Badajoz on 16 October 2026, Juan David Rodríguez García, creator of LearningML, featured it in his talk on positive uses of AI in education, alongside LearningML, as a tool to teach AI by building it rather than only using it. The full text and slides of the talk are available at: https://learningml.org/usos-positivos-ia-educacion-congreso-pide-2026/