ChatGPT lacks kid suitability

While large language models (LLMs) like GPT-3.5 and GPT-4 have made incredible
advancements in natural language processing, they come with significant challenges, particularly in children’s education. These models often generate biased content or produce language that is too complex or inappropriate for younger audiences. Studies have shown that LLMs can perpetuate gender stereotypes, use vocabulary beyond the understanding of children, and even expose users to mature themes.

ChatGPT lacks kid suitability

Moreover, bias in LLMs manifests in various ways - from favoring certain genders in story roles to unintentionally promoting cultural or political biases. For instance, a model might depict boys as heroes while relegating girls to secondary roles. This creates an unfair and skewed educational experience for young learners who may not have the critical thinking skills to recognize and challenge these biases.

Recognizing these limitations, Dima Rubanov and Matthias Neumayer, the co-founders of Oscar Stories (HeyQQ GmbH), have set out to create LORA - a new child-friendly LLM and consumer app.

"In working with AI-generated stories, we found that male children were often depicted as heroes, while female characters were frequently cast as victims. Additionally, the language used, even by GPT-4, is far from child-appropriate. With LORA, we are developing an AI adoption for children that reduces bias and ensures safe, educational experiences," said Dima Rubanov.

LORA: In development, aiming to lead in safe educational AI

LORA is currently in development and aims to offer a superior AI experience for children's education. Unlike existing models, LORA is built on fine-tuned, open-source models, drawing from high-quality, custom-curated datasets created in collaboration with educators. The goal is to outperform current models like GPT-4 or Gemini in both language simplicity and bias reduction.

As the team works toward this vision, the final product will be a mobile app designed to inspire children's interest in STEM fields (science, technology, engineering, and mathematics) through personalized and interactive storytelling. The app will engage children in a way that is fun, accessible, and educational, addressing the gaps left by current AI models.

Future prospects

In the long term, HeyQQ GmbH plans to make LORA's underlying API available to third-party developers in the children’s entertainment and educational sectors. This move will encourage the development of innovative, educational tools and content for young audiences, helping foster a safer, smarter AI landscape.

Matthias Neumayer, co-founder of Oscar Stories, explains: "LORA will be the first trustworthy AI solution for children in the DACH region, designed to reduce bias and deliver child-appropriate language. We aim to bring this technology to life in an engaging mobile app, offering personalized, interactive learning experiences through AI-powered stories."

A team of experts

To support LORA’s development, HeyQQ GmbH has brought on board Marco Mathe, an experienced machine learning expert. The project has received support through the AI Adoption Call from AWS, the Austrian Federal Ministry for Labor and Economy (BMAW), the Fachakademie Sozialpädagogik München Mitte, and is working with TRUSTIFAI for certification as a trustworthy AI solution. 

With LORA, HeyQQ GmbH is setting new standards for AI-supported education, demonstrating their commitment to responsible, child-friendly, and innovative technology solutions.

Key advancements with LORA include:

Age-Appropriate Language: Unlike GPT-3.5 and GPT-4, LORA is fine-tuned to simplify
language based on the child’s age and developmental stage. The language is reviewed by
educators to ensure it is suitable for young readers.

Bias Mitigation: LORA employs a comprehensive approach to identifying and reducing bias, using curated datasets that undergo strict manual verification by educational experts. While AI can never be completely free of bias, LORA takes a proactive approach to minimize its impact.

Personalized Learning through Storytelling: By leveraging children’s interests, LORA integrates educational concepts into engaging narratives. For example, a child interested in construction might learn about gravity from Isaac Newton in a story set on a construction
site.