EMNLP

Lowering temperature should make output safer, not incoherent. Daniel's DeepSeek scripts say otherwise — and the reason may be stranger than he thi...

Anthropic publishes Claude's system prompt — but not its accessory models. What does "leaking" a prompt actually mean?

A tiny 600M parameter model is quietly scanning your ChatGPT tool calls for PII. Here's how that class of guardian model actually works.

Daniel's podcast pipeline works — so why can't he reuse it? Recipes, containers, and Hebrew code-switching TTS.

Heads aren't the Lego bricks of model merging — layers are. Here's what heads really do and how frankenmerges get built.