Concept
Hallucination
Identify plausible claims that are not supported by the available evidence.
Take a method into your next task
Learn
Use concepts, checklists, and guided tutorials to produce a result you can check and reuse.
- What you do
- Choose a concept, reusable method, or step-by-step tutorial.
- Typical time
- 5 minutes for a concept; 10–90 minutes for guided work.
- What you take away
- A checklist, template, or completed work record.
Concepts
Understand the idea, its value, and its boundary.
Start with concepts that are already visible in a reviewed result.
Concept
Source verification
Check that a current, primary source actually supports the nearby claim.
Concept
Human oversight
Place accountable review where errors, privacy, or reversibility matter.
Methods & Guides
Apply a repeatable checklist and keep the record.
Each method connects to published evidence and includes a template you can copy or download.
Judge an AI answer without guessing the model
Review task fit, claims, evidence, uncertainty, and usability before model identity can influence the decision.
- Use when
- You have two or more AI outputs for the same task, or one consequential output that needs a defensible acceptance decision.
- Reusable output
- AI output review record
Used at Richbay
Use the method →Separate observations from inference
Turn sparse results into a decision without presenting labels, causes, or forecasts as observed facts.
- Use when
- You are interpreting pilot metrics, interviews, experiments, research notes, or AI-generated summaries that can easily overstate what the evidence shows.
- Reusable output
- Observation–inference ledger
Richbay tested
Use the method →Audit an AI-generated calculation
Rebuild the arithmetic from source values, preserve intermediate precision, and accept the answer only when the path and result reconcile.
- Use when
- An AI answer contains arithmetic that affects a budget, estimate, report, comparison, or operational decision.
- Reusable output
- Calculation audit worksheet
Richbay tested
Use the method →Tutorials
Produce one checked, reusable result.
Follow a concrete sequence with inputs, human checkpoints, risks, and a final artifact.
Run a reviewed AI output comparison
Define one task, capture controlled outputs, review claims against evidence, and publish a bounded decision record.
- Outcome
- A versioned comparison record and a clearly bounded next experiment.
- Level
- Intermediate
Used at Richbay
Start the tutorial →Build an evidence-bound AI content workflow
Use AI to help research and draft one useful page while a human owns the sources, claims, publication decision, and review date.
- Outcome
- A reviewed page packet with a source ledger, claim checks, disclosure, owner, and next review date.
- Level
- Intermediate
Used at Richbay
Start the tutorial →Cloudflare Workers deployment review guide
Choose the current Next.js deployment path, bind only the services the product needs, and release with a reversible production check.
- Outcome
- A deployment review record covering the chosen adapter, bindings, release checks, and rollback plan for your existing application.
- Level
- Intermediate
Used at Richbay
Start the tutorial →Launch one AI product flow in seven days
Use a seven-session cadence to scope, build, verify, release, and review one complete AI-assisted product flow.
- Outcome
- One usable end-to-end flow, a release record, early observations, and a decision to iterate, package, or stop.
- Level
- Intermediate
Used at Richbay
Start the tutorial →