Capable generalists
Generalist Path
A story-shaped route: short essays first, then the precise cards if you want them.
Read next
Essays · start here
The chatbot passed the test
The model that passes the eval is often not the thing that acts. Seven short essays, with optional tangents.
What this path is for
You have not followed the research literature. You still want a clear way in, in ordinary language.
You will leave knowing:
- Why a green eval on the visible model can miss the thing that actually acts.
- Why objection has to change later behavior, not only the transcript.
- Why copies, slogans, and markets can undo local care.
- Where the precise cards and the book pick up after the stories.
Time: about an hour for the essay thread. Optional tangents open from marked points.
The essays begin at /essay/the-chatbot-passed-the-test/. Appendix D (the fictional hospital gate) is a pointer from the second essay, not the first thing to open.
Read in this order
Cards, chapter toys, experiment lines, and book chapters — curated for this audience. Experiments are sanity checks only; negative results are part of the reading.
The chatbot passed the test
The visible model is often the wrong object.
The hospital is the AI
A deployed tool is the institution plus the model.
They said they heard you
Acknowledgement is not the same as changing tomorrow.
The words stayed
Slogans can survive while tradeoffs reverse.
The copy did not inherit the listening
A successor can keep the words and drop the handles.
The environment picks the winner
Deployment pressure can select against correctability.
A map, not a certificate
Names the problem, and what this project does not claim.