Our Coding Adventure: Using LLMs to Personalise the Narrative of a Tangible Programming Robot for Preschoolers

Abstract

Finding balanced ways to employ Large Language Models (LLMs) in education is a challenge due to inherent risks of poor understanding of the technology and of a susceptible audience. This is particularly so with younger children, who are known to have dificulties with pervasive screen time. Working with a tangible programming robot called Cubetto, we propose an approach to benefit from the capabilities of LLMs by employing such models in the preparation of personalised storytelling, necessary for preschool children to get accustomed to the practice of commanding the robot. We engage in action research to develop an early version of a formalised process to rapidly prototype game stories for Cubetto. Our approach has both reproducible results, because it employs open weight models, and is model-agnostic, because we test it with 5 diferent LLMs. We document on one hand the process, the used materials and prompts, and on the other the learning experience and outcomes. We deem the generation successful for the intended purposes of using the results as a teacher aid. Testing the models on 4 diferent task scenarios, we encounter issues of consistency and hallucinations and document the corresponding evaluation process and attempts (some successful and some not) to overcome these issues. Importantly, the process does not expose children to LLMs directly. Rather, the technology is used to help teachers easily develop personalised narratives on children's preferred topics. We believe our method is adequate for preschool classes and we are planning to further experiment in real-world educational settings.

Keywords

• eol>tangible programming
• preschool education
• LLM storytelling
• open weights models

1. Introduction

Employing Large Language Models (LLMs) in education is not a straightforward task [ 1 ]. It is particularly challenging with younger children, who are already known to have dificulties with pervasive screen time [ 2 ]. At the same time, exactly in the early stages of education, children are very diferent and less used to a standardised educational process. They are less familiar to the abstractions necessary to develop early-stage competences in science, technology, engineering and mathematics (STEM). To illustrate this, already a non-trivial question is “Why should I program a robot to go somewhere, when I can put it there myself ?” and they are not much appreciative of the need to develop computational thinking [ 3 ].

Our contribution is in the context of Cubetto, a physical programming platform: a robot travelling on wheels and a control board with physically-insertable command blocks [ 4 ]. Whereas the creatorprovided teaching materials accompanying Cubetto aim to teach how to use Cubetto, including physical and creative activities for children [ 5 ], they leave it to teachers to elaborate the extensive repetition that would allow children to get accustomed to the robot. We set out from the premise that such getting used to Cubetto and its control is an important prerequisite for its adoption. Thus, we propose an approach to support pre-school teachers in developing the educational storytelling necessary to engage children with the learning activity. This is inspired by the narrative success of AI Dungeon, an interactive platform that engages users with ChatGPT to collaboratively create narrative experiences [ 6 ]. Using LLMs for such a learning activity that is not overly challenging allows for easier discernment of the desired pedagogical features.

2. Background

Computational thinking has been recognised as a key capability that children need to develop in the 21st century and programming skills are a cornerstone to it. Particularly from the 1980s, the practice of turtle programming has stuck as one of the most common onboarding approaches for younger children [ 7, 8 ]. A number of platforms have attempted to make this accessible for even younger audiences by introducing tangibile programming toolkits [ 9, 10, 4 ]. Regardless whether taught on its own or integrated in other subjects, teaching programming skills helps students develop important competences. In early childhood, the process of teaching programming benefits from enrichment with mixture with storytelling activities as a developmentally appropriate delivery approach. A systematic review from 2022 identified a need for the development of educational curricula integrating teaching of programming and the corresponding teacher-training programs [ 3 ]. In a discussion of the barriers to teaching programming to K-12 students, another meta-analysis draws attention to the highly abstract and complex syntax of text-based programming [ 11 ]. The authors indicate particularly the lack of student interest in text-based programming as one of the main challenges and suggest there is a need to improve teaching methods in order to foster student motivation and interest [ 11 ].

Block-based and visual coding tools have emerged as a way to make programming more accessible to younger audiences that might struggle with the traditional text-based approach [ 12 ]. Research indicates that students in elementary schools who learned programming with block-based visual programming tools had better academic achievement than those who learned programming only with traditional text-based programming tools [ 12, 13 ]. One of several toys following a similar rationale, Cubetto has been developed as an attempt to take this further [ 4 ], making block-based programming accessible also to younger children who are especially susceptible to screen time overuse [ 2 ] and the surrounding adults that are increasingly reluctant to have young children engage with digital screens [ 14 ]. Documentation and learning materials around Cubetto in particular [ 4, 5 ] are rich with examples of diferent ways to engage with the robot when it comes to teaching programming skills to small children. This includes diferent physical activities for children, such as imitating and empathising with the robot, story co-creation activities, and creative activities - using drawing, role-play and others.

Research into the motivations to play games [ 15 ] suggests 3 broader groups of motivating categories: achievement, immersion and social motivation. These three categories and their 10 components give broader indication of what it takes to gamify a task. Certainly, challenge and achievement are part of these, and in our particular case these are addressed by the nature of navigating a robot. But the immersion component strongly features storytelling, containing topics such as discovery, role-play, customisation and escapism that could find form through narrative. Shared storytelling among the children could also boost the third category with the components of social motivation, containing the components socialising, relationship and teamwork. Finding ways to include any of these motivational factors reinforces the opportunity to engage children. Our approach is inspired by the narrative success of AI Dungeon, an interactive platform that engages users with ChatGPT to collaboratively create narrative experiences [ 6 ].

3. Method

Here we aim to develop an early version of a formalised approach using rapid prototyping. To this end we engage in action research studying how to prompt LLMs, similar to the way this was previously done regarding the capabilities of a text-to-image transformer [ 16 ]. Such an approach allows us to simultaneously study both the potential of human-AI collaboration for preschool education using current LLMs, and the process of achieving this collaboration. By illustrating the use of five diferent

Domain

Theory

Turtle

Graphics

Pedagogical

Principles

Separation

Engagement

Principles

Immersion

Learning

Activities

Cubetto

models in parallel, we demonstrate that this process could be independent of the specific LLMs of choice.

To guide the development of the intended formalised approach, we turn to a methodological framework for the design of serious games that we have previously developed. We build a process proposed by Davies and Mangan that starts from theory and goes through pedagogical principles to finally arrive at the design of learning activities [ 17 ], which we previously expanded by incorporating both pedagogical and engagement principles in parallel for the development of educational games [ 18 ]. In particular, as pedagogical principles we previously proposed phenomenography, or variation theory, which postulates that in order to efectively understand a phenomenon, a learner needs to experience variation around it. Such variation encompasses three fundamental steps: (i) contrast where one dimension of the phenomenon is played with, while others are maintained fixed; (iii) separation where inversely one dimension is kept fixed; and (iii) fusion where the free interplay between dimensions is experienced [ 19 ]. On the other hand, we also propose engagement principles, based on Yee’s dimensions of achievement, immersion and social motivation [ 15 ]. Within such a framework, we aim to teach basic turtle programming with separation and immersion using Cubetto as illustrated in Figure 1. 3.1. Materials We propose a learning experience with Cubetto, a travelling box-shaped wooden robot. It follows instructions provided by placing physical command blocks on a control board. The standard Cubetto package comes with four commands, but for a preschool audience we use only the three that are about movement (move forward - green, rotate left - yellow and rotate right - red). We also use a slightly enhanced version of Cubetto as visible in Figure 2. In particular, to help children get used to the commands, we use colour markers on top the robot. These markers help associate the commands to the resulting actions.

As a way to address issues of sustained use and reproducibility [ 20 ], for narrative generation our study involves exclusively open weights models. In particular, we consider Google’s Gemma, Meta’s Llama, Mistral, AllenAI’s oLMo and Alibaba’s Qwen. We access these using the llama.cpp platform1 and the GGUF model format. As a way to also ensure accessibility and data privacy, we focus on models that could be run locally on a contemporary laptop without using GPU functionality. These requirements take us to models within the 7-9 billion parameters range and their largest available 3-bit quantisation from HuggingFace2.

Our experiments combine Cubetto with popular commonplace toys. On one hand these additional game subjects are meant as vehicles of inspirations and engagement, on the other as a way for children to overcome the challenge of programming being too abstract, particularly relevant for the age of interest for this study [ 9, 10 ]. Furthermore, this gave us an opportunity to capture how well LLMs are able to handle practical constraints. To illustrate the approach, we choose four toys, popular among preschool children: Barbie dolls, Lego figures, Hot Wheels cars and Brio trains. Each of these becomes a possible value for a parameter in our prompt as could be seen in the next subsection. A particular challenge is presented by the Brio trains subject. These wooden trains are challenging to adapt for two reasons: (i) Cubetto can neither drive along, nor cross the bulky wooden tracks, and (ii) trains follow tracks and thus do not have the afordance of directional movement represented by the rotate left and rotate right command blocks. Whereas this is intuitive to a person, it might not be to a LLM. 3.2. Prompt template and parameters For the composition of the prompt, we take inspiration from narrative structuralist traditions and more specifically the works of Vladimir Propp on morphology of folklore [ 21]. We devise a prompt template to have models generate ideas. In particular we adopt three personalisation parameters: narrative world, subjects and task. The first two are used to set respectively the context and the protagonists so that they could possibly match the children’s current preferences, aspirations and available physical toys. The third one, task, serves to set the objective and possible suggestions for it are derived from Propp’s functions. We chose ones that we consider to relate to a protagonist, as illustrated in Table 1.

In our experimentation we use the combinations provided in the rows of Table 1, but the approach allows recombining them in any permutation or introducing further variants. We provide these parameters to LLMs through the following prompt: 1https://llama-cpp-python.readthedocs.io 2http://hf.co/bartowski/gemma-2-9b-it-GGUF http://hf.co/bartowski/Meta-Llama-3.1-8B-Instruct-GGUF http://hf.co/bartowski/Mistral-7B-Instruct-v0.3-GGUF http://hf.co/bartowski/OLMo-2-1124-7B-Instruct-GGUF http://hf.co/bartowski/Qwen2.5-7B-Instruct-GGUF “Suggest to a teacher a game with the Cubetto tangible programming toy for preschool children. It should be about {narrative world}, involve {subjects} and children should use Cubetto to {task}. Only three Cubetto command blocks should be used: forward, turn left and turn right. The description should be about half a page long. ”

The exact reusable code for this procedure, as well as an overview of the resulting scenarios proposed by LLMs are available in appendix. This, and a full archive of the evolution of the prompts and corresponding generations are available at the online project repository3.

4. Results

Our experiment involved seven rounds of generations to optimise the final results. This count includes only the end-to-end generation where the prompt was combined with all 4 sets of parameters from Table 1.

Generally all tested LLMs work well for this task and provide actionable instructions. Yet due to the issues outlined below, these should not be seen as proof-read and ready to use guides, and rather as creativity prompts for educators instead.

Without being prompted to do so, all models strive to produce structured, self-contained and detailed activity descriptions. These are formatted documents, having length of about one page. The structure of these produced documents is not consistent, even within iterations of the same LLMs. Yet, typical featured sections are scenario topic, objective(s), necessary materials, preparatory setup, instructions or gameplay, claimed learning outcomes, and variations of the proposed activity.

While this type of proposed solutions is generally useful, it leads to a technical problem - generated responses turn out to be larger than the available response token bufer of the LLMs so they get trimmed. On the upside, this typically happens at a point where the suggested task is suficiently clear for a teacher to interpret despite the missing ending. We attempted to resolve the problem of too long responses by adding to the prompt, either requests for half-page answers, or impose a character limit. Gemma, Llama and oLMo appear to consistently ignore this part of the instructions and still produce responses longer than the bufer limit. Mistral and Qwen however, adapt their responses, providing shorter proposals and overcoming the problem.

Inconsistencies and hallucinations are widespread, yet straightforward to overcome by a teacher. In a commonly recurring example of an inconsistency, LLMs pretend to provide a list of required materials, but this is often incomplete, sometimes mentioning some necessary parts of the Cubetto package, but not all (e.g. interchangeably blocks and board, although one cannot function without the other). A typical hallucination is the proposed use of a Cubetto command block that does not exist. We addressed this by adding an explicit list of permissible command blocks, as detailed in Section 3.1. This appears to have worked well and eliminated command-related hallucinations. However, considering that response length is an issue, LLMs consistently repeated the command-restricting requirement - something that we do not see as beneficial to the final task.

Task 1 (enact a pursuit) and task 3 (enact a struggle) could potentially present a challenge, as they could involve two active subjects, whereas models generally propose solutions involving a single Cubetto robot. All models, but Qwen transform task 1 into a rescue mission, i.e. the target does not move. Similarly, for task 3, only Mistral and Qwen do not explicitly transform the task into a rescue mission. Coming back to the question of how many robots are used, the only ones that propose using more than one are oMLo for task 2 (find a treasure) and Qwen for task 3 (enact a struggle). These proposals involve dividing children in groups with a robot for each, even if these are not confronted in an interaction between groups. Thus, the proposed activities are easily adapted to using a single robot. More broadly speaking, all models but oLMo bring up the suggestion to divide children in teams. This makes unprovoked and unnecessary assumptions about the number of children involved and whether this number is suficient and reasonable for a separation in teams. When it comes to the particular challenge of the fourth task, Brio trains, Gemma suggests to “Use Cubetto to guide a train through a Brio track” which appears as dificult to follow through. Other suggestions might lead to the idea of using the Brio tracks as continuous obstacles forming a canyon, which could be seen as a very creative solution. The other LLMs appear to be able to respond abstractly enough not to underline this particular challenge with providing phrases such as “navigating through the Brio tracks” (Llama), “using Brio train tracks to create a winding path” (Mistral). 5. Discussion and Conclusion Our approach installs small models locally and we share all the materials necessary for the reproduction and adoption of this approach. However, should such a process be beyond the technical proficiency of teachers, online access to of-the-shelf models is a completely viable alternative using the same or similar prompts. The method does not disclose any personal information, so there should not be any privacy concerns. However, a subsequent research phase needs to engage with a wider group of children and with pre-school teachers that are independent from the process developers. This would provide evidence about the usefulness of the approach in-the-wild and about perceptions of teachers as direct target audience, and of children as indirect audience. However, the presentation of this method needs to come in the context of the wider learning context of Cubetto, including other introductory activities to the robot [ 4, 5 ].

We insist it is important to emphasise that so far there are no reasons or justifications in favour of exposing children directly to output generated by LLMs. Instead, as we propose here, the technology could be used to help overloaded teachers with rather trivial narratives that would put together elements that these same teachers consider useful to integrate in the learning process. Thus, we pursue a two-fold goal: on one hand to scafold teacher creativity, on the other to not demand from them engagement with the activity that is beyond basic reasonable interaction.

In the process we have encountered cases where LLMs refuse to respond to tasks due to concerns about violence or discrimination. These were rare and circumvented by slight variations in the prompt. However, building on the aforementioned point of no direct exposure of children to LLM output, we believe the premise that censuring responses as a way to protect children is misplaced in educational contexts like ours. It is not up to the model or its authors to decide whether content is appropriate, nor muting conversations is an adequate way of addressing historical controversies. At least for the context in our consideration, we believe that it is teachers that need to take the lead on such decisions. This corresponds to previous research reporting that teachers need to be more involved in educational uses of LLMs [ 1 ].

While definitely a topic of interest, due to the adopted method of this study, our results could not be used as a means of systematic comparison between LLMs. Yet, one cannot help but notice the emergence of consistently diferent behaviour by diferent models. Being diferent solutions to the same next token generation task, the diferences emerge with the exact architectures and the exact training corpora and training procedures. While the research on optimal architectures is ongoing, there is less possibility to study how diferences could be traced back to corpora or training procedures. Among the LLMs studied here, only the creators of oLMo are transparent about both the training process used and put an efort to make the training corpus publicly available [22]. Nevertheless, this one example opens up to the theoretical possibility of studying how responses relate to specific instances in the training data. Declaration on Generative AI The author has not employed any Generative AI tools for the production this article beyond the reported experiments. 50–58

Science Review 6 (2024). doi:10.1162/99608f92.5317da47. [21] V. Propp, Morphology of the Folktale, 2 ed., University of Texas Press, 1968. [22] P. Walsh, L. Soldaini, D. Groeneveld, K. Lo, S. Arora, A. Bhagia, Y. Gu, S. Huang, M. Jordan, N. Lambert, et al., 2 OLMo 2 Furious, 2025. doi:10.48550/arXiv.2501.00656. llama-cpp-python==0.3.9 huggingface-hub==0.31.4 A.2. Script from llama_cpp import Llama The locally executable code that was used for the analysis is provided below. models = [ {"repo": "bartowski/gemma-2-9b-it-GGUF", "file": "gemma-2-9b-it-Q3_K_XL.gguf"}, {"repo": "bartowski/OLMo-2-1124-7B-Instruct-GGUF", "file": "OLMo-2-1124-7B-Instruct-Q3_K_XL.gguf"}, {"repo": "bartowski/Mistral-7B-Instruct-v0.3-GGUF", "file": "Mistral-7B-Instruct-v0.3-Q3_K_L.gguf"}, {"repo": "bartowski/Qwen2.5-7B-Instruct-GGUF", "file": "Qwen2.5-7B-Instruct-Q3_K_XL.gguf"}, {"repo": "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF", "file": "Meta-Llama-3.1-8B-Instruct-Q3_K_XL.gguf"}, for m in models: mname = m["repo"].split("/")[ 1 ].split("-")[0] print(mname) llm = Llama.from_pretrained( repo_id=m["repo"], filename=m["file"], n_gpu_layers=-1, flash_attn=True, ) for p in params: print(p) prompt = prompt_templ.format(**p) msg = [{"role": "user", "content": prompt}] output = llm.create_chat_completion(messages=msg, temperature=0) result = output["choices"][0]["message"]["content"] fname = f’results/{mname}-{p["world"]}-{p["objects"]}-{p["task"]}.txt’ with open(fname, "w") as fout:

fout.write(f"{m[’repo’].split(’/’)[ 1 ]}\n{prompt}\n\n-------------------\n{result}")

B. Outputs

On the next page, the final produced output per model and per task, the last two being indicated in the ifrst paragraph of each minipage. This seen in better detail, and documentation of previous iterations is available at the project repository4. gemma-2-9b-it-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about knights and princes es, involve Barbie dol s and children should use Cubet o to enact a pursuit. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

gemma-2-9b-it-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about pirates, involve Lego figures and children should use Cubet o to find a treasure. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

gemma-2-9b-it-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about superheroes, involve Hot Wheels cars and children should use Cubet o to enact a struggle. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

gemma-2-9b-it-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about Wild West, involve Brio trains and children should use Cubet o to rescue someone.

Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

The Knight’s Quest: A Cubet o Adventure This game combines storytel ing, imaginative play, and basic programming concepts using Cubet o. It’s perfect for preschoolers who are just starting to grasp the idea of giving instructions and seeing them come to life.

Story: Once upon a time, a brave knight set out on a quest to rescue a beautiful princes from a fearsome dragon! The princes is trapped in a tal tower, guarded by the dragon. The knight must navigate a maze to reach the tower and save the princes .

Materials: • Cubet o robot • Cubet o command blocks (forward, turn left, turn right) • Barbie dol s: a knight and a princes • A large play mat or open space • Optional: building blocks, pil ows, or other objects to create obstacles and a “maze” Instructions: 1. Set the Scene: Create a “maze” on the play mat using building blocks, pil ows, or other objects. Place the princes dol at the top of the “maze” and the dragon dol somewhere nearby. 2. Introduce the Knight: Tel the children about the brave knight who is going to rescue the princes . Show them the knight dol and explain that they wil be using Cubet o to guide the knight through the maze. 3. Programming the Knight: • Forward: Explain that the “forward” block wil make Cubet o move forward one step. • Turn Left/Right: Explain that the “turn left” and “turn right” blocks wil make Cubet o change direction. • Chal enge: Have the children work together to program Cubet o to navigate the maze and reach the princes . Encourage them to experiment with dif erent combinations of command blocks. 4. The Rescue: Once Cubet o reaches the princes , have the children celebrate the knight’s succes ! Variations: • Obstacle Course: Add more obstacles to the maze to make it more chal enging. • Dragon Chase: Have the children program Cubet o to avoid the dragon by turning left or right when it gets close. • **Multiple

Ahoy, Mateys! Treasure Hunt with Cubet o! Objective: Guide your pirate crew to find the hidden treasure using Cubet o! Materials: Cubet o’s Wild West Rescue! Theme: Wild West Objective: Use Cubet o to guide a train through a Brio track to rescue a stranded cowboy! Materials: • Cubet o robot • Cubet o command blocks (forward, turn left, turn right) • Brio train track set • Smal cowboy figurine (or any other toy representing the person to be rescued) Setup: 1. Create a Brio track layout resembling a Wild West town. Include obstacles like bridges, tunnels, and maybe even a saloon! 2. Place the cowboy figurine somewhere on the track, stranded and needing rescue.

Instructions: 1. Storytel ing: Begin by set ing the scene. Tel the children a story about a brave cowboy who got lost in the Wild West and is stranded on the tracks. He needs your help to get back to town! 2. Programming: Explain to the children that they wil be using Cubet o to guide the train and rescue the cowboy. Show them the three command blocks: forward, turn left, and turn right. Explain how each block makes Cubet o move in a dif erent direction. 3. Trial and Error: Let the children experiment with dif erent combinations of command blocks to guide the train through the track. Encourage them to try dif erent paths and see what happens. 4. Col aboration: Encourage teamwork! Children can work together to program the train, taking turns adding command blocks and testing their program. 5. Succes ! When the train reaches the cowboy, celebrate their succes ! They have succes ful y rescued the cowboy from the Wild West! Extension Activities: • Obstacle Course: Add more obstacles to the track, chal enging the children to program the train to navigate them. • Dif erent Trains: Use dif erent Brio train cars to represent dif erent characters in the Wild West story. • Storytel ing: Encourage children to create their own Wild West stories and use Cubet o to act them out.

This game al ows children to explore basic programming concepts in a fun and engaging way while developing their problem-solving skil s and creativity. The Wild Meta-Llama-3.1-8B-Instruct-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about knights and princes es, involve Barbie dol s and children should use Cubet o to enact a pursuit. Only three Cubet o command blocks should be used: forward, turn left and turn right.

The description should be about half a page long.

Meta-Llama-3.1-8B-Instruct-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about pirates, involve Lego figures and children should use Cubet o to find a treasure. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

Meta-Llama-3.1-8B-Instruct-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about superheroes, involve Hot Wheels cars and children should use Cubet o to enact a struggle. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

Meta-Llama-3.1-8B-Instruct-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about Wild West, involve Brio trains and children should use Cubet o to rescue someone. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

Game Title: “The Brave Knight’s Rescue” Objective: To create a fun and interactive storytel ing experience using Cubet o, where preschool children can use their problem-solving skil s to help a brave knight rescue a princes from a dragon’s lair.

Materials: Gameplay: • Cubet o programming toy • Barbie dol s (princes and knight) • A simple obstacle course or a designated play area • A smal toy dragon (optional) Game Description: In this exciting game, the children wil take on the role of a brave knight who must rescue a princes from a dragon’s lair. The princes wil be represented by a Barbie dol , and the knight wil also be a Barbie dol . The dragon’s lair wil be a designated play area or an obstacle course. 1. The children wil start by placing the princes Barbie dol in the dragon’s lair, and the knight Barbie dol at the entrance of the lair. 2. The children wil use the Cubet o programming toy to create a sequence of commands to help the knight navigate through the lair and reach the princes . 3. The three Cubet o command blocks that wil be used are: ◦ Forward: to move the knight one step closer to the princes ◦ Turn left: to change direction and avoid obstacles ◦ Turn right: to change direction and reach the princes 4. The children wil need to use their problem-solving skil s to create a sequence of commands that wil help the knight reach the princes without get ing stuck or running into obstacles. 5. As the children create their sequence of commands, they can use the Cubet o to test their solution and see if the knight can reach the princes . 6. If the knight succes ful y reaches the princes , the children can celebrate their succes and try to create a new sequence of commands to make the knight’s journey even more chal enging.

Tips and Variations: • To make the game more chal enging, you can add obstacles such as smal blocks or toys that the knight

Treasure Hunt Adventure with Cubet o Objective: Help the pirates find the hidden treasure using Cubet o’s simple programming commands.

Materials: • Cubet o programming toy • Lego pirate figures • A treasure chest (or a smal box) • A pirate-themed obstacle course (e.g., a smal island, a bridge, or a cave) Instructions: 1. Set up the pirate-themed obstacle course in the clas room or play area. Place the treasure chest at the end of the course. 2. Introduce the Cubet o programming toy and explain that the children wil use it to help the pirates find the treasure. 3. Demonstrate how to use the Cubet o command blocks: forward, turn left, and turn right. Explain that the children wil need to program the Cubet o to navigate through the obstacle course to reach the treasure. 4. Divide the children into smal groups and give each group a Lego pirate figure. Explain that they wil need to work together to program the Cubet o to find the treasure. 5. Chal enge the children to program the Cubet o to reach the treasure using the three command blocks: forward, turn left, and turn right. Encourage them to experiment and try dif erent combinations of commands. 6. As the children program the Cubet o, ask them to describe their plan and explain why they chose a particular sequence of commands. 7. Once the Cubet o reaches the treasure, congratulate the children on their succes and ask them to reflect on what they learned about programming and problem-solving.

Variations: • To make it more chal enging, add obstacles or “pirate traps” that the Cubet o needs to avoid. • Encourage the children to create their own pirate-themed obstacle course and chal enge their peers to find the treasure. • Use the Cubet o to create a pirate-themed story, where the children take turns programming the Cubet o to navigate through a series of chal enges and obstacles.

As es ment: • Observe the children’s ability to use the Cubet o command blocks to solve a problem

Superhero Rescue Mis ion with Cubet o and Hot Wheels Objective: To create a fun and interactive game that encourages preschool children to use problem-solving skil s and critical thinking while learning the basics of programming with Cubet o.

Game Description: Gameplay: In this exciting game, children wil become superheroes tasked with rescuing their friends from the clutches of the evil vil ain, “The Blocker.” The Blocker has set up a series of obstacles using Hot Wheels cars and blocks, and it’s up to our young superheroes to navigate through the chal enges and reach their friends. 1. Set up the game area with a series of Hot Wheels cars and blocks to create a chal enging obstacle course. 2. Divide the children into teams of 2-3 superheroes. 3. Each team wil receive a Cubet o robot and a set of command blocks (forward, turn left, and turn right). 4. The objective is for the superheroes to program their Cubet o to navigate through the obstacle course and reach their friends, who are trapped behind the Hot Wheels cars. 5. However, The Blocker has set up a series of chal enges to hinder the superheroes’ progres . For example, a Hot Wheels car might be placed on a steep incline, or a block might be used to block the Cubet o’s path. 6. The superheroes must use their problem-solving skil s and critical thinking to program their Cubet o to overcome these chal enges. 7. The team that succes ful y rescues their friends and reaches the final destination wins.

Cubet o Command Blocks: • Forward: Move the Cubet o forward one block. • Turn Left: Turn the Cubet o left by 90 degrees. • Turn Right: Turn the Cubet o right by 90 degrees.

Tips and Variations: • To make the game more chal enging, you can add more Hot Wheels cars and blocks to the obstacle course. • You can also add time limits or set up a series of chal enges that the superheroes must complete before reaching their friends. • Enc

Wild West Rescue Mis ion Objective: Help the brave Cubet o cowboy rescue the lost cowgirl from the Wild West town.

Materials: • Cubet o programming toy • Brio trains and tracks • Wild West themed props (e.g., cowboy hats, bandannas, cacti) • A smal toy cowgirl figure Setup: 1. Create a Wild West town using Brio tracks and props. Set up a smal town with a main street, a saloon, and a sherif ’s of ice. 2. Place the toy cowgirl figure in a hard-to-reach location, such as a high platform or a smal cave. 3. Explain to the children that the cowgirl is in trouble and needs to be rescued.

Gameplay: 1. Give each child a Cubet o programming toy and explain the three basic commands: forward, turn left, and turn right. 2. Chal enge the children to program their Cubet o cowboy to reach the cowgirl using the three basic commands. 3. Encourage the children to think creatively and use problem-solving skil s to overcome obstacles, such as navigating through the Brio tracks or avoiding “bandits” (other toys or props). 4. As the children program their Cubet o, ask them to describe their plan and explain why they chose a particular sequence of commands. 5. Once the children have succes ful y rescued the cowgirl, ask them to reflect on their proces and discus what they learned about problem-solving and programming.

Variations: • To add an extra level of chal enge, you can introduce “bandits” or “obstacles” that the Cubet o cowboy must avoid or overcome. • Encourage the children to create their own Wild West-themed props and incorporate them into the game. • Consider set ing up a time limit or a series of chal enges for the children to complete, such as rescuing the cowgirl from dif erent locations or navigating through a maze.

As es ment: • Observe the children’s problem-solving skil s and ability to think Mistral-7B-Instruct-v0.3-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about knights and princes es, involve Barbie dol s and children should use Cubet o to enact a pursuit. Only three Cubet o command blocks should be used: forward, turn left and turn right.

The description should be about half a page long.

Mistral-7B-Instruct-v0.3-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about pirates, involve Lego figures and children should use Cubet o to find a treasure. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

Mistral-7B-Instruct-v0.3-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about superheroes, involve Hot Wheels cars and children should use Cubet o to enact a struggle. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

Mistral-7B-Instruct-v0.3-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about Wild West, involve Brio trains and children should use Cubet o to rescue someone. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

Title: “Knightly Pursuit: A Cubet o Adventure” In this engaging and imaginative game, preschool children wil embark on a thril ing knightly pursuit using the Cubet o tangible programming toy. The story revolves around a brave knight and a daring princes , both Barbie dol s, who are separated in a magical kingdom.

The children’s task is to guide the knight, represented by Cubet o, on a quest to find the princes . The journey wil be fil ed with obstacles, hidden treasures, and friendly creatures.

To control Cubet o, the children wil use three simple command blocks: “Forward,” “Turn Left,” and “Turn Right.” The knight wil move forward to traverse the kingdom, turning left or right to navigate around obstacles, find hidden paths, or approach the princes .

The game encourages problem-solving, spatial reasoning, and storytel ing. As the children guide Cubet o, they wil learn the basics of programming while immersed in a captivating knightly tale. The game’s open-ended nature al ows for endles replayability, as children can create their unique paths and adventures.

This game not only introduces young minds to the world of coding but also fosters creativity, col aboration, and a love for storytel ing. The combination of physical play with the Cubet o toy and the imaginative knightly pursuit wil make learning a fun and memorable experience for the children.

Title: Pirate Treasure Hunt with Cubet o and Lego Pirates Objective: In this engaging and interactive game, preschool children wil use the Cubet o tangible programming toy to navigate a pirate-themed Lego island in search of a hidden treasure.

Materials: - Cubet o programming robot - Cubet o block set (forward, turn left, turn right) - Lego pirate figures and acces ories - Lego island landscape (including a treasure chest) Instructions: 1. Set up a Lego pirate island landscape, complete with a treasure chest hidden somewhere on the island. 2. Place the Cubet o programming robot at the starting point on the island. 3. Introduce the children to the Cubet o blocks (forward, turn left, turn right) and explain that they wil use these blocks to guide Cubet o on his treasure hunt. 4. Divide the children into teams and provide each team with a set of Lego pirate figures and acces ories. 5. Instruct the children to program Cubet o using the given blocks to navigate the island, find the treasure, and bring it back to their team’s starting point. 6. Encourage teamwork, problem-solving, and critical thinking as the children work together to solve the treasure hunt puzzle. 7. The team that succes ful y finds the treasure and brings it back to their starting point first wins the game.

This Pirate Treasure Hunt game combines the fun of pirate adventures with the educational benefits of programming, fostering creativity, col aboration, and problem-solving skil s in preschool children.

Title: Superhero Car Chase with Cubet o and Hot Wheels Objective: In this engaging and interactive game, preschool children wil use the Cubet o tangible programming toy to create a thril ing superhero car chase scenario. This activity wil help children develop their problem-solving skil s, spatial awarenes , and storytel ing abilities while having fun.

Materials: - Cubet o programming robot - Cubet o block set (forward, turn left, turn right) - Hot Wheels cars and tracks - Superhero figurines Instructions: 1. Set up a Hot Wheels track with various twists, turns, and obstacles to create an exciting car chase environment. 2. Introduce superhero figurines to represent the characters in the story. 3. Explain to the children that they wil be using the Cubet o programming blocks to control the superhero’s vehicle and help them escape from the vil ain’s car.

Gameplay: 1. Place the Cubet o robot at the starting point of the Hot Wheels track. 2. Children wil use the forward block to make Cubet o move along the track. 3. To navigate turns and obstacles, children wil use the turn left and turn right blocks. 4. The objective is for the superhero to reach the finish line before the vil ain catches up. 5. Encourage children to strategize and use problem-solving skil s to create the most ef icient path for their superhero. 6. After each run, discus the strategies used and brainstorm ways to improve them for the next round.

This game not only fosters children’s creativity and critical thinking but also helps them understand the basics of programming and problem-solving in a fun and engaging way. By incorporating superheroes and Hot Wheels, the game appeals to children’s natural sense of adventure and competition, making learning a joyful experience.

Title: Cubet o’s Wild West Rescue Mis ion Objective: In this engaging and educational game, preschool children wil use the Cubet o tangible programming toy to navigate a Wild West adventure, rescuing a friend from a perilous train predicament.

Materials: Cubet o programming robot, Brio train tracks, as orted Wild West themed props (e.g., cacti, tumbleweeds, cowboy hats), and a friend in need (represented by a smal figure).

Instructions:

OLMo-2-1124-7B-Instruct-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about Wild West, involve Brio trains and children should use Cubet o to rescue someone. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long. 1. Preparation: ◦ Characters: Use Barbie dol s to represent the knight (Sir Valor) and the princes (Princes Seraphina). ◦ Cubet o Setup: Ar ange the Cubet o robot in the starting position at the foot of a smal hil (made from paper or blocks) that represents the entrance to the enchanted forest. ◦ Map: Create a simple map using paper or a whiteboard, il ustrating the path Sir Valor must take.

Include landmarks like the enchanted forest, the crystal river, and the mountain where the princes is held captive. 2. Programming Commands: ◦ Forward: The knight moves forward along the path. ◦ Turn Left: The knight turns left to explore dif erent paths or avoid obstacles. ◦ Turn Right: The knight turns right to approach the next landmark or chal enge. 3. The Quest: ◦ Starting Point: The knight begins at the foot of the hil , facing the entrance to the enchanted forest. ◦ First Chal enge: The knight must turn left and move forward to cros the enchanted forest. ◦ Second Chal enge: After exiting the forest, the knight encounters a crystal river. To cros , the knight must turn right

Set ing the Scene: Transform your clas room into a vibrant pirate cove. Ar ange the Cubet o play area to resemble a sun-drenched island, complete with a sandy beach, a smal pirate ship, and a treasure chest. Scat er Lego pirate figures around the island, each with a distinct role: the captain, first mate, par ot, and crew members.

The treasure chest is placed at the end of a winding path, guarded by a friendly sea creature made from a large Lego block. 1. The Setup: ◦ Begin by explaining the story to your students: Cubet o, the brave robot pirate, has received a map leading to the legendary treasure hidden on the island. The map is encoded in a sequence of commands using Cubet o’s blocks. ◦ Introduce the Lego pirate figures and the treasure chest as key elements of the adventure. 2. The Commands: ◦ Teach the children the three basic commands: forward (to move Cubet o along the path), turn left (to navigate around obstacles), and turn right (to change direction). ◦ Demonstrate how to string these commands together to create a simple program that guides Cubet o from the starting point to the treasure. 3. The Adventure Begins: ◦ Divide the clas into smal groups, each with their own Cubet o robot and set of commands. ◦ Encourage them to work together to decode the map (a simple sequence of commands you provide) and guide Cubet o to the treasure. ◦ As they play, circulate and of er guidance, helping them troubleshoot any chal enges they encounter. 4. **Interactive Story 1. Setup: Place the Barbie dol s on the play mat, with the knight and princes in dif erent starting positions.

The play mat can be decorated with a medieval castle, a forest, or a magical garden to enhance the theme. 2. Objective: The children wil program Cubet o to fol ow the knight as he pursues the princes . The goal is to create a path that leads Cubet o from the knight’s starting position to the princes ’s position, with the knight trying to catch up. 3. Gameplay: Each child wil take turns programming Cubet o using the forward, turn left, and turn right blocks. They can add obstacles or treasures along the path to make the pursuit more exciting. For example, they might program Cubet o to turn left around a tree or turn right to avoid a dragon. 4. Rules: The knight (Cubet o) must fol ow the path created by the children’s commands. If Cubet o reaches the princes ’s position, the knight wins! If the princes manages to escape by changing the path, she wins. 5. Variations: To keep the game interesting, children can switch roles, with one child programming Cubet o as the knight and another as the princes . They can also add dif erent chal enges, such as a time limit or a requirement to col ect specific items along the way.

This game encourages children to think creatively, practice problem-solving skil s, and develop an understanding of basic programming concepts through a fun and engaging nar ative.

Qwen2.5-7B-Instruct-GGUF Suggest to a teacher a game with the Cubet o tangible programming toy for preschool children. It should be about pirates, involve Lego figures and children should use Cubet o to find a treasure. Only three Cubet o command blocks should be used: forward, turn left and turn right. The description should be about half a page long.

Certainly! Here’s a fun and engaging game idea for preschool children using the Cubet o tangible programming toy, themed around pirates and treasure hunting: Pirate Treasure Hunt Adventure Objective: Guide Cubet o, the friendly wooden robot, through a pirate-themed map to find the hidden treasure using only three command blocks: forward, turn left, and turn right.

Materials: - Cubet o - Programming board - Command blocks (forward, turn left, turn right) - Lego figures (pirates, treasure chest, map pieces) - A large play mat or a flat surface Setup: 1. Create a pirate-themed map using Lego figures and other materials. Place a treasure chest at the end of the map. 2. Scat er Lego pirates and other obstacles along the path to the treasure. 3. Place Cubet o at the starting point of the map.

Gameplay: 1. Divide the children into smal groups and give each group a set of command blocks. 2. Explain that Cubet o is a brave pirate captain on a quest to find the treasure. 3. Each group must program Cubet o to navigate through the map and find the treasure using only the three command blocks. 4. Encourage the children to work together, discus their strategies, and test dif erent combinations of commands. 5. Once a group succes ful y guides Cubet o to the treasure, they can place a Lego pirate figure on the treasure chest to celebrate their succes . 6. Rotate the groups and repeat the game, al owing each group to have a turn as the captain of Cubet o.

References

[1] E. Harvey , A. Koenecke , R. F. Kizilcec , "Don't Forget the Teachers": Towards an Educator-Centered Understanding of Harms from Large Language Models in Education , in: Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , ACM, Yokohama Japan , 2025 , pp. 1 - 19 . doi: 10 .1145/3706598.3713210.

[2] C. Wang , H. Qian , H. Li , D. Wu, The status quo, contributors, consequences and models of digital overuse/problematic use in preschoolers: A scoping review , Frontiers in Psychology 14 ( 2023 ). doi: 10 .3389/fpsyg. 2023 . 1049102 .

[3] E. Macrides , O. Miliou , C. Angeli , Programming in early childhood education: A systematic review , Int. J. of Child-Computer Interaction 32 ( 2022 ). doi: 10 .1016/j.ijcci. 2021 . 100396 .

[4] Solid Labs , Beginning computer programming for kids: An introductory guide to computational thinking and coding for kids aged 3-6 years old , PRIMO Toys , 2017 . URL: https://www.primotoys. com/wp-content/uploads/2017/09/Ebook-PrimoToys_ final - 1 .pdf.

[5] Solid Labs , Coding with Cubetto - Unit 1, Reception, Ages 4 to 5 , UK National Curriculum, PRIMO Toys , 2017 . URL: https://primotoys.com/education/resources/.

[6] M. Hua , R. Raley , Playing With Unicorns: AI Dungeon and Citizen NLP , Digital Humanities Quarterly 14 ( 2020 ). URL: https://www.proquest.com/docview/2553526112.

[7] S. Grover , R. Pea , Computational Thinking in K-12: A Review of the State of the Field , Educational Researcher 42 ( 2013 ) 38 - 43 . doi: 10 .3102/0013189X12463051.

[8] T. S. McNerney , From turtles to Tangible Programming Bricks: explorations in physical language design , Personal and Ubiquitous Computing 8 ( 2004 ) 326 - 337 . doi: 10 .1007/s00779-004-0295-6.

[9] L. D. Rodić , A. Granić , Tangible interfaces in early years' education: a systematic review , Personal and Ubiquitous Computing 26 ( 2022 ) 39 - 77 . doi: 10 .1007/s00779-021-01556-x.

[10] S. Papavlasopoulou , M. N. Giannakos , L. Jaccheri, Reviewing the afordances of tangible programming languages: Implications for design and practice , in: 2017 IEEE Global Engineering Education Conference (EDUCON) , 2017 , pp. 1811 - 1816 . doi: 10 .1109/EDUCON. 2017 . 7943096 .

[11] L. Sun , L. Zhou , Does text-based programming improve K-12 students' CT skills? Evidence from a meta-analysis and synthesis of qualitative data in educational contexts , Thinking Skills and Creativity 49 ( 2023 ) 101340 . doi: 10 .1016/j.tsc. 2023 . 101340 .

[12] Y. Hu , C.-H. Chen , C.-Y. Su, Exploring the Efectiveness and Moderators of Block-Based Visual Programming on Student Learning: A Meta-Analysis , Journal of Educational Computing Research 58 ( 2021 ) 1467 - 1493 . doi: 10 .1177/0735633120945935.

[13] L. Hu , Programming and 21st century skill development in K-12 schools: A multidimensional meta-analysis , J. of Computer Assisted Learning 40 ( 2024 ) 610 - 636 . doi: 10 .1111/jcal.12904.

[14] A. Gentleman , ' the crux of all evil': what happened to the first city that tried to ban smartphones for under-14s?, The Guardian ( 2025 ). URL: https://www.theguardian.com/technology/2025/may/07/ the-crux -of-all-evil-what-happened-to-the-first-city-that-tried-to-ban-smartphones-for-under-14s.

[15] N. Yee , Motivations of Play in MMORPGs, in: DiGRA 2005 Conference, 2005 , p. 46 . URL: http://www.nickyee.com/daedalus/motivations.pdf.

[16] M. Ruskov , Grimm in Wonderland: Prompt Engineering with Midjourney to Illustrate Fairytales, in: B. Alessia , F. Alex , F. Stefano , M. Stefano , R. Domenico (Eds.), Proceedings of the 19th Conference on Information and Research Science Connecting to Digital and Library Science , volume 3365 of CEUR Workshop Proceedings , 2023 , pp. 180 - 191 . URL: https://ceur-ws. org/ Vol- 3365 /#paper6.

[17] P. I. Davies , J. Mangan , Embedding Threshold Concepts: from theory to pedagogical principles to learning activities , in: R. Land , J. H. Meyer , J. Smith (Eds.), Threshold Concepts within the Disciplines , Brill, Rotterdam, 2008 , pp. 37 - 50 . doi: 10 .1163/9789460911477_ 004 .

[18] M. Ruskov , Employing Variation in the Object of Learning for the Design-based Development of Serious Games that Support Learning of Conditional Knowledge , PhD Thesis , University College London, 2014 . URL: https://discovery.ucl.ac.uk/id/eprint/1457529/.

[19] F. Marton , Necessary Conditions of Learning, Taylor and Francis, Hoboken, 2014 .

[20] L. Chen , M. Zaharia , J. Zou , How Is ChatGPT's Behavior Changing Over Time? , Harvard Data