# How a Content Creator Built a Personal AI Model After Being Banned Twice by OpenAI
A popular content creator and digital personality has shared his unusual journey of building a small artificial intelligence model designed to run on personal computers, a project that drew the attention — and the bans — of one of the industry’s biggest players.
## A Personal AI Project With a Catchy Name
The project, called Ajax, is a fine-tuned AI model built to operate locally on consumer hardware. Rather than relying on cloud-based services that require subscriptions and send user data to remote servers, Ajax is designed to work entirely on a person’s own machine — handling tasks like sorting email inboxes and browsing the web independently.
The model is reportedly built on Alibaba’s open-source Qwen framework and is expected to contain roughly 9 billion parameters, the adjustable numerical values that determine an AI model’s capabilities. While larger models with trillions of parameters dominate headlines, Ajax represents the growing trend toward smaller, more accessible AI that ordinary users can run at home without paying recurring fees.
## The Road to Ajax: Two Account Bans
The creator, known for his massive online following, says the path to building Ajax involved some contentious interactions with OpenAI, the company behind ChatGPT.
His goal was to use OpenAI’s flagship reasoning model to generate training data for Ajax through a process called distillation — essentially training a smaller model by having it learn from the outputs of a larger, more capable one. During this work, he says his OpenAI account was suspended twice.
The first ban occurred after he began extracting the model’s internal reasoning traces — the hidden “scratchpad” text that AI systems generate while working through problems. The second ban came when he attempted to use the model’s outputs as seed data, the foundational examples a new AI learns from. OpenAI’s terms of service prohibit using its outputs to develop competing models, according to the creator’s account.
He disputed both suspensions, and the account was restored after the first ban. However, the second suspension followed shortly after he resumed work using the same approach.
## Removing AI Guardrails
Once training was underway, the creator faced another challenge: ensuring Ajax would not refuse to answer user requests the way mainstream AI assistants sometimes do. To accomplish this, he used Heretic, an open-source tool designed to strip away a model’s built-in refusal mechanisms.
The technique, sometimes called “abliteration,” works by analyzing how the model responds to both harmful and harmless prompts, identifying the pattern behind its refusals, and removing that pattern. The result is an AI that consistently complies with user requests without the safety guardrails found in commercial alternatives.
The creator acknowledged the trade-off. Ajax is not perfect — he described it as having sustained “a little brain damage” from the procedure — but said it successfully completes everyday tasks like inbox management and web browsing about 90 percent of the time. He also drew a firm line, stating that he chose not to remove refusals related to causing harm to others or to oneself.
## Training Method and Ongoing Development
Ajax is still a work in progress. The creator is currently training it using a reinforcement-based method called GRPO, where the model attempts a given task multiple times and learns from the attempts that produce the best results.
He described Ajax as a “version one” product, noting that there are still steps to complete, including rerunning ablation tests, compressing the model through a process called quantization, and running benchmarks to ensure nothing broke during the aggressive training modifications.
## The Countdown to Release
The creator originally set a release date for Ajax, posting a countdown timer on the project’s download page. The timer was scheduled to reach zero at 08:25 Japan time on October 3. At this point, the countdown is no longer visible on the page, though the model itself is not yet downloadable. The creator indicated the model will be released once he determines it is ready.
## Why Small Models Matter
His advocacy for smaller, locally-run models centers on two main arguments: cost and energy consumption. By his estimates, a single copy of a rumored trillion-parameter frontier model would require approximately 27 personal computers and the electricity equivalent of about 150 homes to operate. A model like Ajax, by contrast, can run on a standard home computer, making powerful AI accessible without the infrastructure demands of massive data centers.
—
## Frequently Asked Questions
**What is Ajax?**
Ajax is a small AI model, approximately 9 billion parameters in size, built to run locally on personal computers. It is designed to handle everyday tasks like email management and web browsing without requiring cloud connectivity or subscription fees.
**Why was the creator banned by OpenAI?**
The creator’s OpenAI account was suspended twice. The first occurred while he was attempting to extract reasoning traces from OpenAI’s flagship model for use as training data. The second happened when he tried to use the model’s outputs as seed data for Ajax, which OpenAI’s terms of service prohibit.
**What is distillation in AI?**
Distillation is a technique where a smaller model is trained using the outputs of a larger, more capable model. Think of it as a student studying the work of an expert to learn a skill. The smaller model inherits some of the larger model’s capabilities while being more efficient to run.
**What is abliteration?**
Abliteration is a process that removes an AI model’s built-in refusal mechanisms, allowing it to comply with a wider range of user requests. This is done using specialized tools that identify and eliminate the patterns that cause the model to decline certain prompts.
**Why run AI models locally?**
Running AI locally keeps user data on personal hardware rather than sending it to company servers. It also eliminates subscription costs and reduces the energy infrastructure required compared to running massive cloud-based models.
**Is Ajax available to download now?**
No. While the creator originally set a countdown timer for a release date, the model is still in development and not yet available for download. The release will happen once the creator determines the model is ready.
**What safeguards does the creator mention?**
The creator states that he chose not to remove refusals related to causing harm to other people or to oneself. He also notes that his lawyer advised him to clarify that Ajax is not designed to provide dangerous or actionable instructions.
—
## Conclusion
The story of Ajax highlights the tension between accessibility and control in the AI industry. As creators find ways to build capable models that run on consumer hardware, questions about terms of service, model safety, and the democratization of AI technology continue to grow louder. Whether Ajax succeeds as a practical tool for everyday users or serves as a case study in the challenges of independent AI development, it represents a compelling chapter in the ongoing story of who gets to build and use artificial intelligence.
Thank you for reading
