THE FORMAT
Temporary laboratories disguised as conversations.
The useful interview begins where the prepared biography stops.
We want to know what the guest actually has to decide, measure, build, repair, infer, defend, or explain. That means moving quickly from credentials into mechanisms: what constraints matter, what the public gets wrong, what failed, what evidence changed the model, and what still refuses to fit.
The conversation should leave behind durable artifacts. Definitions. Diagrams. Source trails. Failure cases. Corrections. Open questions. A reader should be able to return months later and find more than the memory that an interesting person once said interesting things.
QUESTION SET / CORE
The questions we keep bringing back.
- What do intelligent outsiders consistently misunderstand about your work?
- What constraint dominates the problem but rarely appears in public explanations?
- What failed the first time you tried to solve it?
- What evidence changed your mind about something important?
- Which metric looks useful until you understand the system better?
- What can a simulation or model not tell you?
- Where does the system fail gracefully, and where does it fail catastrophically?
- What would make your current conclusion wrong?
- What piece of working knowledge is difficult to transfer without firsthand experience?
- What unresolved question should we hand to the next expert?
ENGINEERING DESK
Machines are very good at exposing vague answers.
For engineers, fabricators, roboticists, electronics specialists, aerospace people, mechanics, and field operators, the interview shifts toward failure modes and decision tradeoffs. We want dimensions, margins, service conditions, instrumentation, tooling, maintenance, sourcing, heat, vibration, contamination, tolerances, access, and what happened after the elegant drawing met the ugly machine.
What did revision two know that revision one did not?
That single question often opens the entire build history.
SCIENCE DESK
Claims need a path back to observation.
For researchers, the interview should distinguish model, measurement, inference, uncertainty, and interpretation. What does the instrument actually detect? What transformation turns that signal into a quantity? Which assumptions enter the pipeline? How robust is the result to a different model? Where are the dominant uncertainties?
The goal is not to perform peer review on a bar stool. It is to reveal enough of the reasoning chain that a curious reader can understand why a result deserves confidence and where that confidence stops.
AI + SOFTWARE DESK
Benchmarks, edge cases, and the cost of being wrong.
Software interviews tend to become abstract too quickly, so we force them back toward behavior. What does the system do under load? What does it log? What can be rolled back? How is failure detected? Which benchmark correlates with actual usefulness? Where does automation remove toil, and where does it merely conceal a decision?
For AI, we also ask what would count as evidence that the system is not improving. A model that can only be validated by people who already believe in it has wandered out of engineering and into religion, which already has plenty of departments.
HUMAN SYSTEMS DESK
Ask what keeps people functional when theory meets Tuesday.
For clinicians, therapists, researchers, social workers, addiction specialists, crisis workers, and relationship professionals, the interview shifts toward warning signs, recurring failure patterns, thresholds, recovery, autonomy, and what people can actually do before a situation becomes harder to reverse.
We also interrogate the profession itself: what gets oversimplified, what smart people rationalize especially well, where online mental-health language distorts useful concepts, what AI can and cannot safely contribute, and what the field still does not understand.
CORRECTIONS
The interview does not become sacred after publication.
Guests can clarify factual errors. Editors can append corrections. Sources can be added. New evidence can change the durable reference derived from a conversation. The record should show meaningful changes rather than silently rewriting history into a cleaner shape.
The archive will preserve a distinction between what the guest said, what Cyberdelia independently verified, what remains inference, and what changed later. That separation is tedious right up until it becomes essential.