AI Experiments (Containers, Blog, Conversations, Free Money)

Purpose

After integrating ChatGPT into the Words app, the next natural question was: what else can this automate? These four experiments were rapid-fire probes — each took a different domain (containerization, content generation, discourse simulation, financial trading) and asked "can AI do this usefully?"

Stack

What I Learned

• free-money-1 was the most interesting experiment — a trading bot that used AI to make buy/sell decisions. The name says everything about the naivety of the premise. Trading bots face a fundamental adversarial environment: every edge gets arbitraged away, markets are efficient enough that consistent alpha is extraordinarily hard to find, and the difference between backtesting profit and live trading profit is usually the difference between profit and loss. The experiment taught healthy skepticism about AI trading claims.
• blog-1 explored automated content generation — using AI to produce blog posts or log entries from structured data. The output was coherent but generic. AI-generated content without a human voice reads like a Wikipedia article rewritten by a marketing intern. The lesson: AI is excellent at drafting, structuring, and expanding — but the voice, the opinion, the specific insight that makes content worth reading still requires a human in the loop. These project blog posts are the evolved version of that experiment.
• conversations-1 was the most conceptually ambitious: pull trending topics from Google Trends, then simulate adversarial discussions between AI personas who debate the topic from opposing viewpoints. The goal was to rapidly surface the strongest arguments on both sides of any trending issue. The output was interesting but suffered from AI's tendency to be balanced and reasonable rather than genuinely adversarial. Real debates have heat. AI debates have diplomacy.
• containers-1 is in this list and I genuinely do not remember what it was for. It may have been a Docker/containerization experiment, or it may have been something else entirely. Not every project in a fast-experimentation phase survives with context. The commit messages, if they exist, are the only archaeology available.
• The cluster pattern — four experiments in rapid succession, none of them deep — is characteristic of early AI adoption. The technology was new enough that the smart move was breadth: try everything, see what sticks. Going deep on the wrong application wastes more time than going shallow on four applications and identifying the one worth pursuing.

Key Insights

• The trading bot experiment (free-money-1) taught the most transferable lesson: if something seems like free money, you are the product, not the customer. In trading, the "free money" is the spread — the difference between what you think the market will do and what it actually does. In AI, the "free money" is the assumption that AI output is correct without verification. Both require skepticism and risk management. The XRPL trader project in the Potatuhs ecosystem inherited this skepticism.
• Automated content generation has a quality ceiling without human direction. The blog-1 experiment showed that AI can produce volume but not voice. The project blog you are reading right now is the answer: a human (me) provides the stories, opinions, and context. An AI (Claude Code) provides structure, synthesis, and articulation. Neither alone produces the output. Together, the ceiling lifts.
• Simulating adversarial discussion with AI (conversations-1) is a precursor to what is now called "red teaming" in AI development — using AI to stress-test ideas by generating opposing arguments. The technique has matured since this experiment. Modern approaches use multiple AI agents with different system prompts, fine-tuned for different rhetorical styles. The concept was right. The execution was too early.
• Rapid-fire experimentation is the correct strategy when a new technology arrives. You cannot know which application will matter until you try several. The cost of each experiment was low (days, not months). The information value was high (clear signal on what works and what does not). Four small experiments taught more about AI's capabilities and limitations than one large project would have.