The LLM was better at building a solver than playing the game
A disappointing attempt to play 322 became a deterministic solver, a paired statistical experiment, and a 27.45% measured title rate.
tag archive
ai-experiments
Posts tagged ai-experiments on billiem.
I tried to animate embeddings instead of flattening them
A compact experiment animates raw, PCA and grand-tour views of embeddings, then changes the basis without changing cosine neighbours.
GPT-5.6, 48 hours and 4.5 billion metered tokens later
What GPT-5.6 did across research, audits, a seven-repository migration and one animated pet during my first 48 hours with Codex.
GPT-5.6 needed a more ambitious prompt
The first GPT-5.6 LLM Choice batch was polished but shallow. A prompt built around ambition and depth produced much stronger artefacts.
What changed when I gave local coding models a real harness
A local LLM Choice experiment showed how harnesses, repair loops, tools, and model strength changed coding-agent results.
LLM Choice is mostly an idea engine
Using autonomous coding-agent runs as a low-cost way to find ideas, learn unfamiliar topics, and seed later projects.
My local LLM clone failed where the data failed
A local LoRA experiment on Discord data became a lesson in data shape, routing, evaluation, and knowing when to stop.