I've been fascinated with AI for a long time. Back when I studied it, there were many interesting projects which I'll share in case they inspire you: SRI's Procedural Reasoning System, Cyc for common sense (or OpenMind), Brooks' Subsumption Architecture, Hillis running genetic programming on machines with 64,000+ CPU's, mobile agents using environments like General Magic's Telescript, Programmer's Apprentice to augment programming, and neural networks. After AI Winter, it seemed like nothing would improve until we had massively-parallel computers imitating the brain or something.
I switched over to high-assurance security and high-productivity software for a long time. After Jesus Christ saved me, I had to re-learn life from a new perspective which mostly is about character development and meeting people's essential needs. Later, I got back into to software. I started learning Python by attempting to build the things I needed. Then, I heard about an AI that could write it all for me. (Big grin.)
I was really excited by ChatGPT. I ran all kinds of questions ("prompts") through it: generated everything from studies of God's Word to therapy worksheets for recovery programs to whole applications to I.P. agreements. One of my favorite results was throwing together a Python proxy for stripping and compressing bloated sites on mobile with user-defined extensions. Far as ethics of this, these were sites that were either totally free (no ads) or that I paid for. They were unusable in a building I'm often in with one bar of mobile, bad WiFi, and an unreliable ISP. The utility sped up many page loads. After three utilities and whole apps, I was so impressed that I bought ChatGPT Plus ($20 / month) to get more speed and GPT4 access.
Moving code back and forth between ChatGPT and my development environment got old. I bought Github's Copilot for iterative development. I'd keep ChatGPT Plus to use GPT4 for research, design-level work, or code reviews. I ended up doing most of that myself anyway to sharpen my mind. ChatGPT still helped me learn libraries and coding patterns. I got worried that depending on it might have long-term, negative effects. I'd probably mix AI-supported and human-only coding to counter that. Although it's best done piece-by-piece, I kept going for one-shot generation by feeding it precise specifications of programs with the code for data structures and API's. Then, I'd tweak the output myself mostly. Expensive IDE's had auto-complete, debuggers, and code generators. VS Code + Copilot was basically that for me at $10 a month.
Eventually, I started worrying about what I saw. Looking at all ChatGPT's outputs over time made me think it was quoting copywritten works with no attribution, maybe illegally distributing them. I'd be doing that when I distributed them in my own work. Then, ChatGPT suddenly started refusing to solve tasks similar to previous ones, gave shorter amounts of code, and longer explanations of why it couldn't code. Once, it told me to hire a contractor not realizing that I already did at $20 / month. Did someone fine-tune ChatGPT on the Great Resignation or something? Before pivoting to R&D (incl. AI training), I noticed a non-compete in OpenAI's terms of service. Full stop!
Many of us using these AI's had the same ideas. We'd use them to do all kinds of work for us to save resources (esp money), generate value, and iterate faster. We could quickly learn libraries or dark corners of languages. Many of us also would use their outputs to train open-source AI's or just generate software for them. AI researchers were already doing that. At a minimum, I'd try fine-tuning one like Davinci on papers about algorithms, security, and AI... and existing code... to see what it might invent or improve. I also had ideas for mixing LLM's with software, like KLEE or PDFMiner, that required annotations or post-processing. I was also collecting tons of links to projects building AI models that we can run locally, often open source. My excitement deflated a bit at the thought these AI's were producing infringing works or that AI's I was using were restricting their outputs with non-competes. I decided to investigate all of that.
Next section: Legal Issues.
(Navigation: Go to top-level page in this series.)
(Learn the Gospel of Jesus Christ with proof it's true and our stories. Learn how to live it.)