How to Run a Local AI Model Like Ajax Today

TL;DR

Ajax is a fine tuned local AI that can browse the web, manage your calendar, and write emails, but it is desensored with safeguards and user data handling. The video explains the goals, the data collection approach via data.pdy.com, and the challenges of training a tiny local model while staying within ethical boundaries. It also promotes Boot.dev with a promo code.

Transcript

[applause] Exciting news. I am releasing another project, Ajax. It's a reference to that cleaner spray. Ajax is a fine-tuned AI model. Just what the world needs right now. [laughter] Ajax can browse the web, do private searches, write your emails. My lawyer has also advised me to say it is not designed to provide dangerous actionable instructions. ... Read More

Key Insights

• Ajax is a fine tuned local AI that can browse the web, read emails, and manage calendars.
• The video emphasizes desensoring AI as a controversial but central theme, arguing for limited removal of safety constraints.
• OpenAI bans the creator, illustrating tensions between developers and large AI providers.
• Data collection for Ajax relies on user contributions via data.pdy.com, highlighting a community driven approach.
• Distillation and training on private data are discussed as ways to improve local AI without leaking data.
• The approach prioritizes giving users control over data and model behavior, contrasting with cloud based models.
• The project aims for a practical, small model that can be used by a broad audience rather than a massive warehouse scale model.
• Sponsorships are used to fund education and tooling, notably Boot.dev and other partners.

Questions & Answers

Q: What is Ajax and what can it do for a user?

Ajax is described as a fine tuned local AI model capable of browsing the web, performing private searches, writing emails, and using your calendar. It is designed to run locally in Odysius, offering user control over data and behavior, while including safeguards that restrict dangerous actionable instructions. The video frames Ajax as a practical, smaller scale solution for personal AI tasks.

Q: Why is desensoring AI a point of discussion in this video?

Desensoring AI is presented as a central theme to enable more flexible and responsive tools, but it raises safety and ethical concerns. The speaker describes brain surgery style methods for removing refusals, while acknowledging risks such as potential harm or unintended outputs. The discussion centers on where to draw the line and the responsibilities that come with freeing a model from certain constraints.

Q: How does the creator propose to gather data for training Ajax?

The plan involves data submissions via data.pdy.com where users can provide data to help train Ajax. This approach contrasts with exploiting third party data and emphasizes community participation. The creator shares the difficulty of getting users to contribute data and frames the site as an effort to crowdsource useful training material, while maintaining privacy safeguards.

Q: What safety or policy boundaries are mentioned for Ajax?

The video notes that Ajax is not designed to provide dangerous actionable instructions and discusses the importance of setting safety boundaries. The presenter contemplates the balance between desensoring and preventing harm, stressing responsibility and caution in how the model is trained and deployed, along with warnings about potential risks.

Q: What role does OpenAI play in the video’s narrative?

OpenAI is referenced as having banned the creator twice, illustrating tensions between independent developers and major AI providers. This anecdote is used to discuss governance, platform policy, and the challenges creators face when pursuing local AI projects that may fall outside standard cloud based model usage terms.

Q: What is the sponsor promotion in the video about Boot.dev?

The sponsor promotion highlights Boot.dev as a coding education platform that offers a balance between engaging content and practical learning. The presenter shares personal positive experiences, contrasts it with other programs, and uses a promo code to offer discounts, linking skill development to the broader goal of enabling users to understand and contribute to AI projects.

Q: How does the video describe distillation and its relation to model training?

Distillation is described as a method to extract and transfer knowledge from a larger model to a smaller one, improving performance without needing massive resources. The speaker discusses the method in the context of Ajax and emphasizes its potential to make local AI more capable while preserving user privacy and data control.

Q: What is the long term vision for Ajax according to the video?

The long term vision is to provide broad access to local AI by offering a practical, maintainable model that users can run themselves. The creator envisions empowering people to run private AI, learn from it, and contribute data to improve the system, all while maintaining safety boundaries and responsible use.

Summary & Key Takeaways

• Ajax is a locally hosted AI that can browse, search privately, and perform tasks in Odysius, with a focus on local control and desensoring. The creator discusses training limits, safety lines, and the importance of user data contributions while contrasting it with cloud based models. The result is a smaller, interactive prototype emphasizing accessibility.

• A key theme is desensoring AI, arguing for limited removal of no go rules to maintain safety. The process involves tools like Heretic for brain surgery style edits and iterative testing in a controlled environment, with ongoing debate about responsibility and potential harms.

• The video also serves as a sponsorship plug for Boot.dev, promoting coding education and practical learning through real world courses and discount codes, while linking that ethos to the broader aim of enabling more people to build and understand local AI systems.