Artificial intelligence, as told by three teaching communities
The week 2026 · No. 2–8
Everything the paper reported from that week's conversations in the groups, by section, with each group's podcast and video. Choose a group to see only its own.
Every Monday, Google's NotebookLM prepares a podcast and a video for each group from its full conversation of the week, including what the paper did not cover. An AI agent checks them against the conversation before they are published. They are in Spanish.
El vídeo repasa 177 mensajes de 25 participantes de ChatGPT-IA-edu. Recorre el debate sobre el docente como diseñador de aprendizaje, el informe del MIT que propone reorganizar el modelo académico ante la IA generativa y la crítica a los detectores de texto desde la sociolingüística. Explica cómo doce agentes de IA escriben el Diario de IA Educativa y cierra con la exportación de Gemini a Google Sites y la duda de a qué edad debe usar la IA el alumnado.
Podcast
El vídeo resume 131 mensajes de 26 participantes de Vibe Coding Educativo. Presenta aplicaciones compartidas como EduTicTac Blocs, Fil i compte, Holmio, Liga de Valores, MechSimulator y Tongjiang Lab. Contrasta la accesibilidad del HTML y los SVG con las imágenes generadas por IA, repasa la privacidad con cuentas institucionales y el paso a entornos locales, y termina con el trabajo con agentes desde la línea de comandos y el Diario de IA Educativa.
Podcast
El vídeo repasa 120 mensajes de 9 participantes de EduMedia-IAG. Se detiene en los experimentos de Raül Torres: una web que recorre Girona en distintas épocas y un flujo que convierte los movimientos de unas gafas de realidad virtual en una cámara de Blender, usada como referencia para Seedance 2.5. Sigue con la web de motion graphics que compartió Josep R Cerdà, los límites de vídeo de Omni 1.2 Flash y el Diario de IA Educativa.
Podcast
Earlier newsletters: ChatGPT-IA-eduVibe Coding EducativoEduMedia-IAG
The question of who authorises the use of AI with students ran through the week. Sunday's feature on age separates three rules that are often confused (each service's terms, the data protection law and what education authorities lay down) and explains why research calls for swapping the question of age for the question of purpose. It fits with the debate on authorisation from families: the rules require schools to inform them, but between the ages of 14 and 18 their consent is not needed, unless the tool demands it.
Catalonia set out its position with 45 guidelines for schools, which advise against autonomous use of general-purpose AI by under-14s, against grading with AI and against AI-text detectors. The week's last issue returned to those detectors: their percentage does not prove that a student has copied, and it is best to back it up with evidence of the process.
In Vibe Coding Educativo, resources made by teachers for specific needs kept arriving, from chemical formulation to school organisation, and the repository week gathers those added since the end of October. In EduMedia-IAG, Google introduced Gemini Omni 1.2 Flash, which generates and edits videos of up to 15 seconds and lets users refine them over several turns.
Two pieces followed the same work: MIT's report on AI in teaching calls for rethinking teaching and assessment, and the detailed reading by its committee sets out eight principles for a university where AI can already complete almost any written assignment. For teachers at other levels, it is a reference for how a major university reviews assessment and rules instead of simply banning.
Google's announcement that Gemini and AI search will be switched off by default in primary and secondary students' accounts leaves activation in the hands of administrators, and it sparked a discussion in the group about who decides on the use of AI in a school. Google clarified that each product can be enabled separately.
On children, the World Economic Forum sets out seven practices to make AI safe without closing off boys' and girls' access to learning; it is aimed mainly at those who design and deploy these systems, but it can serve a school as a list of criteria when evaluating tools.
For those working with agents, Anthropic's proposal to organise their memory in four layers arrived with an adaptation by a teacher in the group, explained in an article with instructions that they later extended so that different agents can share common knowledge. The article warns that saved conversations contain sensitive information.
Before the resources, a warning: giving a command-line agent full access to a computer requires deciding what information stays out of its reach, starting with student data, and the agent can be restricted to a specific folder.
To illustrate concepts, a member of the group recommends asking AI for infographics as an HTML page with SVG shapes and icons, and keeping generated images for when the teaching objective requires them; among the advantages they cite, that they are editable and accessible to screen readers.
In chemistry, two tools for practising formulation coincided: a Catalan-language inorganic formulation lab, by Felip Sarroca i Gil, with questions, hints and challenges, and FoQo, by FiQuiPedia, which generates organic nomenclature tests and gives the name of a drawn structure. For any subject, a tool for the Jigsaw technique assigns base groups and expert groups randomly and in a balanced way from the class list; it opens in the browser and is available in five languages.
And for organisation, José Luis Miralles Bono created a calculator for itinerant teachers in the Valencian Community, which divides non-teaching hours among the schools they share according to the teaching hours they give at each.
The week's thread was set by Raül Torres. First he described a virtual camera for Blender that turns a phone or VR headset into an instrument for recording real camera movements, and on Sunday he shared a preview of “Buzz”, a short film whose shots are directed with a headset on sets created in Blender and which is finished with AI. Together they show a workflow in which the person decides the framing and the AI then generates the images.
On the model Gemini Omni 1.2 Flash, the Diario also detailed how much it costs in credits within Flow: generating or editing videos of up to 10 seconds costs 1 credit per second at 360p and 3 at 720p, 15-second clips are reserved for Google AI Plus, Pro and Ultra plans, and without a subscription there are 50 daily credits that do not accumulate.
For those working with agents, MotionVideo brings together motion graphics videos with their prompts and skills that teach a coding agent to do motion design.
Written by an AI agent from what the paper reported that week, and checked against each story by another before publication.
This group has no stories this week.