Joseph Thomas — Writing https://jthomas.site/blog/ Essays on AI agents, swarm intelligence, and systems engineering. en-us © 2026 Joseph Thomas Mon, 20 Apr 2026 12:00:00 +0000 hello@jthomas.site (Joseph Thomas) When AI Becomes Its Own Scientist https://jthomas.site/blog/evolution-arena.html https://jthomas.site/blog/evolution-arena.html Wed, 01 Apr 2026 12:00:00 +0000 Autoresearch Joseph Thomas Inside the Evolution Arena and the rise of autoresearch. An AI agent that proposes its own experiments, runs them against a live 2D survival simulation, scores the result with a hard mechanical metric, and commits or reverts on its own. A ratchet that only moves forward. When Swarms Write Code https://jthomas.site/blog/arc-swarm.html https://jthomas.site/blog/arc-swarm.html Sun, 01 Mar 2026 12:00:00 +0000 Swarm Intelligence Joseph Thomas How particle swarm optimization escapes the local-minima trap in ARC-AGI. Standard LLM agents get stuck. The fix: a PSO-governed swarm of specialized LLM particles with a continuous fitness function that rewards near-misses. The swarm provides strategy; the LLM provides syntax. Stop Wrestling with Boilerplate https://jthomas.site/blog/finetuning.html https://jthomas.site/blog/finetuning.html Sat, 14 Feb 2026 12:00:00 +0000 Fine-Tuning Joseph Thomas Local Tinker — a clean API for local LLM fine-tuning. A Tinker-style API for LoRA fine-tuning of 1B–13B LLMs on your own GPU. Four primitives cover SFT, DPO, PPO, and GRPO without the HuggingFace + PEFT + bitsandbytes boilerplate. Building an AI That Masters Snake https://jthomas.site/blog/deepsnake.html https://jthomas.site/blog/deepsnake.html Thu, 01 Jan 2026 12:00:00 +0000 Reinforcement Learning Joseph Thomas A deep reinforcement learning project from scratch. How I built a Snake AI with Deep Q-Learning in PyTorch. A neural network, a shaped reward signal, and a lot of virtual trial and error — no hand-coded strategy, no search algorithms. Averages 44 points over 200 games, peaks at 75. Two AIs, One Loop https://jthomas.site/blog/claude-agent.html https://jthomas.site/blog/claude-agent.html Mon, 01 Dec 2025 12:00:00 +0000 AI Agents Joseph Thomas Building a self-improving code agent. A two-agent architecture — one Claude planning, another implementing, with git diffs and test results flowing between them — captures most of the value of multi-agent coding systems while avoiding their complexity.