Tinkering
Here are some things I've been tinkering with:
2026
BenchmarksExperiments in what vision models can see. Explore the reports, methods, and reproducible source data.
Talk to FlyChat with GPT-Fly: a GPT-2 whose weights are wired from a real fruit fly connectome, running entirely in your browser.
FontBenchA visual font identification benchmark evaluating multimodal LLMs on fine-grained typographic discernment.view repo · view post
BorderBenchA visual benchmark testing how multimodal LLMs perceive borders, corners, and shadows.view repo
Agent Trace ViewerVisualize agent traces in Vercel AI SDK format.
KayaA scripting language for agents.
2025
2024
Hawke(taken down) An AI street stall you could haggle with and buy real products from. I took it off the web because I got tired of drop-shipping people Salty Licorice.
2023
LOLLMA stand-up comedy podcast written and performed by AI, using a fine-tuned GPT-3.