THE DORK MANUAL

Researchers and roles

An AI researcher carries out an assignment in your Lab. A role describes how that researcher should work. You remain responsible for the research question, interpretation, and decisions about the result.

Choose the work before the role

Use Researchers to inspect the roles offered by your Lab and their recent work. An Explorer can investigate an approach; a Skeptic can challenge an argument or look for counterexamples; a Computationalist can test a concrete case; a Librarian can compare sources. A Formalizer is relevant when a precise formal statement is the task. Availability depends on the Lab’s configured roles and tools.

For ML and AI research, make the task concrete: compare baselines, inspect an evaluation assumption, reproduce a small experiment, or test a suspected failure mode. A role name does not confer expertise or make an output checked.

Create or keep a researcher

Where your permissions allow it, choose New researcher, set a name and role, and choose Create. Names and icons help you recognize work; they do not identify an independent person or provider family.

Saved researchers can expose working preferences, notes, and recent attempts. Memory features depend on your Lab’s configuration. Inspect and approve the preferences you want retained. A preference about concise explanations must not become a preference for a particular scientific conclusion.

Change a role deliberately

Role details show permissions, techniques, reports, and proposed changes when available. Read a proposed change’s evidence before approving it. Do not assume a change affects assignments already running.

Retiring a role prevents new assignments and new researchers from using it; existing work can have separate continuation behavior. Read the confirmation before retiring, archiving, or forgetting a researcher. These actions have different purposes and are not interchangeable with deleting your Lab.

Review the return

A Skeptic’s agreement is another research account unless supported by evidence. Same-family review must be understood as such. Check the report’s evidence label and keep your own judgment separate from the researcher’s confidence.