Tech News — Latest News

Why a Small Transformer Can't Copy a Word It Hasn't Seen

I have a small transformer that turns a one-line spec into a working web app. It is 11.9M parameters, 6 layers, d=384, trained on 4,176 generated pr…

measurement codegeneration llm tokenization

Code: Megapixel99/capture-the-flag I built a capture-the-flag arena where language models attack and defend each other's containers. The replays the…

llm measurement finetuning security

Four-bit training quantizes every number to one of 16 values. NVFP4's menu is {0, ±0.5, ±1, ±1.5, ±2, ±3, ±4, ±6} , with one scale factor per block…

llm quantization training measurement

I asked models to extract facts with provenance two different ways. Naive: return {answer, quote} , insist the quote is verbatim, then string-match…

llm extraction provenance measurement

Code: Megapixel99/capture-the-flag In April I ran five games of an AI capture-the-flag tournament between five small open-weight models (1.0B to 2.5…

llm measurement security benchmarks