miniblog.

An AI benchmark website that tries to run comparable benchmarks regularly to discover when LLM performance is degrading:

Playing with GitHub actions today, and I realise how much testing services have become a commodity. I really appreciate GitHub's organised, collapsible UI for output. The featureset seems comparable to alternatives though. (The idea of an ecosystem of actions is tempting!)

I do most of my computing on a desktop I built for gaming in 2011, with minimal upgrades since. Looking at components now, it's striking how comparable parts get better performance but still put out less heat!