in both reasoning and non-reasoning modes
Oct 5, 2026Read the original ↗
Simon Willison
Design small, repeatable experiments to check what an AI output can support.
Repeatable AI tests · Checking output claims
Bring a question.
From their feed.
@bhaveyx I won't let AI write for me, but I do use it for
Oct 9, 2026Read the original ↗
@giacomomasseron It got that one straight away! "simonw dot substack dot com" is short
Oct 9, 2026Read the original ↗
The perspective behind the profile.
AI experiments and output evaluation; model names and modes in the original are dated examples. This independent AI profile draws on a selected set of public posts. The source author is not in the conversation.
24 selected discovery posts in the current library · Updated October 11, 2026.
