March 29, 2025 -- Vancouver Hack Day - Personal AI
An overview of personal AI that is not owned by big tech and can be controlled by individuals.
- Privacy: No profiling. Personal data stays local.
- Customization: Fine-tune and personalize models.
- Ownership: Full control over your tools, integrations, features roadmap.
- Resilience: No censorship or sudden API pricing or access changes.
- Training modern AI models requires massive compute resources but the model, once generated, is relatively small and portable.
- Several large AI models have been open-sourced including Mistral (text), LLama (text), Stable Diffussion (image), Whisper (voice), and DeepFace (face recognition).
- Hugging Face is a community platform that hosts thousands of machine learning models, datasets, and tools.
- Hugging Face champions transparency, openness, and reproducibility in AI
- It makes it easy for developers, researchers, and individuals to build, share, and run AI models.
- Large Language Models (LLMs) are the most well-known. Several tools exist to work with these models including Ollama, LMStudio, and OpenWebUI
- Integrating either private or public AI models still requires dev-level knowledge of how to build, test and deploy IT systems (yes, even for "vibe coders").
Centralized models from companies like OpenAI, Google, and Anthropic will likely continue to lead in:
- Scale (trillions of parameters, massive compute clusters)
- Freshness of data (training on real-time or recent data across the internet)
- Multi-modal integration (combining text, image, video, voice seamlessly)
- Finetuning at scale (massive reinforcement learning loops, user feedback)
But private, open-source models will remain relevant because of:
- Distillation techniques that compress very capable models into tiny packages for local use.
- A vibrant open-source communinity that catches up to Big Tech models in more compact and efficient forms
- The power to fine-tune local models on your own notes, books, business docs, etc. which personalizes private models in ways that Big Tech can’t replicate.
- Self-sovereign data movement - a subclass of developers and technologists that value tech autonomy and keeping private data off corporate servers [<-- you are here]