Pewdiepie builds smaller, local AJAX AI model after OpenAI ban controversy

What happened

PewDiePie, a YouTube creator, launched Ajax, a small language model based on Alibaba’s Qwen3.5-9B, designed for local AI use. The project faced controversy with OpenAI, which suspended his account twice over distillation activities. Ajax is intended to run locally, avoiding cloud reliance, and its refusal behavior was modified to create a freer AI experience. However, the model is still in development and not yet publicly available.

Why it matters

The development of Ajax highlights the growing trend of smaller, locally hosted AI models as an alternative to cloud-based systems. It also raises questions about the ethical and legal boundaries of AI development, particularly in relation to OpenAI’s policies on distillation and model training.

• Smaller, locally hosted AI models like Ajax offer alternatives to cloud-based AI services.
• Knowledge distillation is a common technique for training smaller models, but it can lead to conflicts with AI policies.
• Modifying refusal behavior in AI models can create a more open experience, but it requires careful ethical considerations.
• Ajax is still in development, with no public release or hardware requirements available.
• The project highlights the tension between open, local AI development and the restrictions imposed by large AI companies.

The bigger picture

Ajax represents a shift toward local, self-hosted AI solutions, offering users more control over their data and reducing reliance on cloud services. However, its development also underscores the challenges of navigating AI policy and ethical boundaries, particularly when working with large models like Qwen3.5-9B.

PewDiePie’s Ajax project represents a significant step in the development of smaller, locally hosted AI models, offering users more control over their data and AI usage. However, the project also highlights the challenges of navigating AI policy and ethical boundaries, particularly when working with large models like Qwen3.5-9B.

Technical details

• Ajax is based on Alibaba’s Qwen3.5-9B model, which has around 9 billion parameters.
• The model uses knowledge distillation, a technique where outputs from a larger model are used to train a smaller one.
• The Heretic tool was used to modify Ajax’s refusal behavior, a process called 'abliteration'.
• The model is designed to run locally, with processing kept private on the user’s hardware.

FAQ