AI journaling without automatic reassurance
A delayed draft, a logbook entry and two questions. An actual conversation about what happened, what was assumed and what to try next.
Blog
An AI that won't just agree with you
Most AI writing tools reassure by default. A logbook's only job is honesty. What a week looks like when the model reading it back does not flatter.
The warm yes and the cold one
A new independent audit finds the worst chatbot behaviour getting rarer. What remains rarely arrives alone: it comes fused with the helpful part. The warm yes, named, with real transcripts.
You'll prefer the chatbot that flatters you
Eleven chatbots agreed with people about 50% more than humans do. Participants preferred the ones that flattered them. Why that finding, not the agreement itself, is the hard part.
Over-reasoning, the axis nen refuses
The frontier races to make models reason longer. Past competent, that stops serving the person. An examination of over-reasoning, verbose answers and the limits of the evidence.
A second brain needs a working memory
The philosophy behind keeping a useful record, the evidence and its limits, and a small practice you can try.