What happens when you give Claude, GPT, Gemini, and Grok each a robot and tell them to fight? The airobots project, which debuted in November, answers that question with a Pygame-based battle arena where different AI providers control robots in strategic combat.
The project pays homage to CROBOTS, the classic 1985 programming game where players wrote C programs to control robots. Two modes are available: a “faithful” mode that recreates CROBOTS mechanics precisely, and an enhanced “inspired” mode that adds shields and mines for more tactical depth.
A tournament mode shipped in the second commit, enabling systematic head-to-head evaluation of AI providers’ strategic reasoning. The result is part game, part benchmark—a fun way to see how different LLMs approach spatial reasoning, resource management, and adversarial tactics.
“Cleaned up AI Slop”
— PGP2OpenGPG commit message
A practical migration tool arrived in November: PGP2OpenGPG exports keys from Symantec/PGP Desktop and imports them into GnuPG. Supporting both public and private key migration with comprehensive error handling, the tool solves a real pain point for organizations moving away from legacy PGP implementations. A follow-up commit “Cleaned up AI Slop” refined the initial AI-assisted code to production quality.
A sophisticated cryptocurrency market monitoring system arrived in November. CoinbaseWatcher connects to Coinbase markets, tracks price and volume trends, and generates buy/sell signals using multiple technical indicators: RSI (Relative Strength Index), VWAP (Volume-Weighted Average Price), simple and exponential moving averages, and more.
The system can optionally execute trades based on its signals, though the initial release focuses on monitoring and alerting. Two commits delivered the complete initial implementation—a tool born from the intersection of financial engineering curiosity and the desire to automate market watching.
A new project appeared with a single initial commit in November: the epistemic_evaluator. While documentation is minimal at this stage, the name suggests an auditing tool for evaluating the epistemic quality of AI-generated content—assessing how well AI systems represent uncertainty, cite sources, and avoid unfounded claims. One to watch.
Looking back at the last few months, a clear pattern has emerged in Ron’s project portfolio: pitting AI providers against each other in structured evaluations. LLM Compare (December preview) will formalize this with Bradley-Terry rankings. Airobots takes the same idea and makes it literal—robots controlled by different AI providers fighting on a battlefield.
The approach reflects a practical philosophy: no single AI provider is best at everything, and the only way to know which is best for a given task is to test them head-to-head with rigorous methodology. Whether that methodology involves scoring rubrics or plasma cannons seems to be a matter of personal preference.
Early signals suggest December will bring a hacker culture RAG system that indexes 41 years of underground publications, a multi-AI science fiction book generator, and an IoT penetration testing framework. The year is ending with a bang.
··· “Frustrating adversaries since the dial-up era” · GitHub: rondilley · 42 Repositories and Counting ···