Today's post is a bit off the beaten path from what you might be used to. I'm sharing a personal experiment with AI, not for hype or groundbreaking revelations, but as a straightforward recount of my experiences. This is for those with a decent grasp of AI, looking to see what happens when you mix language modeling with a slice of personal history.
Starting Point
This whole thing started with a simple idea: what if I used my own digital footprints to train an AI model? I'm talking about a heap of my writings, digital conversations with myself on a rubber ducky account, and random thoughts. The end goal was to see if I could make a chatbot that kind of talks and thinks like me. It's a bit strange, feeding a bunch of personal data and private musings into a data set, but curiosity got the better of me. This exploration isn't for the AI experts, but more for those who've dabbled in AI and are curious about seeing their digital reflection.
The Technical Side
The journey began with a blog post about fine-tuning LLMs (Language Learning Models) with cloud services. My toolkit included Runpod for hosting the training, Axolotl for the training process, HuggingFace for managing my data, and Weights and Biases for monitoring. But this wasn't just about copying a guide. It was about understanding the process from the ground up – trying different hardware setups, understanding resource demands, and getting hands-on with the whole training process.
Data Privacy: A Big Deal
When your experiment involves personal data, keeping it secure is crucial. I managed all data preparations locally and used HuggingFace's private repositories for safe storage. This was about more than just security – it was a nod to the importance of ethical AI practices.
The Model Training Experience
Training the model turned out to be a dive into my past. The data spanned a good part of the last decade, capturing different periods of my life. When I tested the model using text-generation-ui, it was strange and somewhat intriguing to see it mimic my past thoughts and styles. It was more than just a digital echo; it was a window into my past self.
Ethics and Self-Reflection
Building a digital twin wasn't just a technical exercise. It pushed me into some deep thinking about ethics and identity. What does it mean to create a digital entity that mirrors your thoughts? It's a bit of a philosophical maze, thinking about the rights and respects of a digital entity that's essentially you but not you.
Education Over Restriction
In this data-driven age, understanding the ramifications of AI is more important than ever. I lean towards the idea that education, rather than stringent rules, is the key to handling data responsibly. Can you imagine where we'd be if Github kept you from committing code that seemed like it might maybe sorta kinda be ethically unsound? Or worse, if those restrictions were in place without a clear/immediate path forward for resolving false positives? This project made me think about the broader implications of AI, privacy, and identity – not just for me, but for everyone in the digital age.
What's Next on This Path
The experiment is far from over. I'm now looking into how this technology can be applied to predictive response modeling and other practical uses. But more than anything, this journey has been a lesson in technology, ethics, and personal introspection.
Join the Conversation
If AI, ethics, or just tech talk is your thing, feel free to join the discussion on my discord. Let's explore these topics together.
Wrapping Up
I hope it gives you a glimpse into the more personal side of tech experimentation. Keep an eye out for more updates, and let's continue exploring AI possibilities together.
