¿Solo quieres estudiar? Entra sin inscribirte
El curso IA generativa y pensamiento dialéctico tiene una edición de lectura abierta con sus 22 lecciones. No necesitas una cuenta para consultarlas. Solo debes registrarte y matricularte en Aula Clara si quieres presentar los cuestionarios, registrar tu progreso y obtener certificado.
1.2 · From a Prompt to a Working System: Sources, Tools and Controls
Goal: understand how the same question changes when an assistant can retrieve sources, analyze files, calculate values, write code or act through connected systems.
A model without data cannot observe your business
Imagine asking, “Where do we lose customers?” A model without access to business records can propose plausible mechanisms but cannot know your actual losses. Give it a fictional table of visits, orders and repeat purchases, and it can test more specific hypotheses. With a calculation tool it may produce reproducible ratios. With a connected data system it may retrieve live records, subject to permissions. These are different levels of evidence.
Five operational modes
- Idea generation: draft explanations and alternatives; no measurement yet.
- Working from a source: inspect a document or dataset, recording its origin and coverage.
- Computational analysis: calculate from inputs; inspect code, formulas, denominators and units.
- Artifact building: generate a calculator, chart or simulation; independently test what it actually implements.
- External action: modify an external system only with clear authorization, scoped permissions and human review.
Two reports can look identical
One report contains a polished graph of completely invented monthly sales. Another produces the same graph from a dated and authorized CSV with disclosed filters. Their visual quality is equal. Their evidentiary status is radically different. Yet the second still requires checking duplicate records, missing values and unit definitions.
A working agreement with an assistant
Our decision is [goal]. The only available and authorized data are [sources]. Before analyzing, state what you can actually inspect or execute. Separate observation, calculation, inference and recommendation. For any code or retrieved source, record the input, transformation and output. Do not take actions on external systems without explicit approval.
Exercise
Run the coffee-shop case three ways: without data, with a small fictional order log, and with a proposed simulator. Compare which conclusions become defensible at each stage. Identify where the tool merely described a calculation instead of performing it. Deliverable: a traceability map from input to operation to conclusion, remaining uncertainty and responsible reviewer.
Para registrar avances, presentar los cuestionarios y recibir un certificado debes inscribirte en el curso.
Inscribirme para guardar mi progreso