95% de la gente AÚN escribe MAL sus prompts en ChatGPT-5

Understanding GPT-5: Key Changes and Hacks

Introduction to GPT-5's Limitations

• The speaker discusses common frustrations users face when interacting with GPT-5, highlighting that many feel let down by its responses.

• Despite being OpenAI's most advanced model, there are fundamental architectural changes that affect how prompts yield results.

Fundamental Changes in GPT-5

Consolidation of Models

• GPT has consolidated all its models into a single interface, reducing the variety of options available to users compared to previous versions.

• An "invisible router" now determines whether a prompt is processed by a quick response model or an advanced one, complicating user experience.

Performance Variability

• The effectiveness of the invisible router can lead to unpredictable outcomes; users may receive either rapid or advanced responses without clear indication.

Understanding Prompting Techniques

Shift in Instruction Handling

• Unlike its predecessors, GPT-5 excels at following precise instructions but struggles with vague or ambiguous tasks.

Importance of Orientation Phrases

• The speaker introduces five tricks for maximizing the utility of prompts in GPT-5. The first hack involves using orientation phrases to trigger deeper reasoning processes within the model.

Effective Prompting Strategies

Using Specific Phrasing

• By adding specific phrases at the end of prompts (e.g., "think deeply about this"), users can encourage the system to engage more thoroughly with their queries.

Avoiding Ambiguity

• Users should avoid vague expressions like "this is important," as they do not provide clear guidance on what is expected from the AI.

Controlling Output Quality

Verbosity Parameter

• A new parameter called verbosity allows users to specify how concise or detailed they want responses from GPT. This feature helps save time and improve response quality.

Understanding Verbosity in Task Management

Types of Verbosity

• Low Verbosity: Ideal for urgent tasks requiring clear, concise answers. Users can request essential information in 100 words or less using markdown for clarity.

• Medium Verbosity: Useful when some context is needed, such as explaining sales trends in a meeting. A request for three to five paragraphs strikes a balance between detail and brevity.

• High Verbosity: Necessary for detailed learning tasks. Users can ask for comprehensive breakdowns of around 600 words, leveraging GPT-5's improved word count capabilities.

Prompt Optimization Techniques

• Identifying Prompt Issues: Recognizing that the root problem may lie within the user's prompt itself is crucial. This realization can lead to better outcomes.

• Metaprompting Technique: Utilizing GPT-5 as a personal prompt engineer helps refine and structure prompts effectively, enhancing their quality.

Implementing Metaprompting

• Definition of Metaprompting: An advanced technique where one uses a language model to generate and optimize other prompts instead of just writing them directly.

• Creating Effective Prompts: By instructing GPT-5 to critique and improve prompts, users can eliminate ambiguities and ensure clarity in requests.

Structuring Prompts Effectively

• Understanding Machine Interpretation: The way humans perceive structured text differs from machines; what seems organized to us may appear as unformatted text to AI.

• Using Structured Data Formats: Employing formats like JSON or XML allows machines to interpret data correctly, minimizing errors during processing.

Example of Structured vs. Unstructured Prompts

• Unstructured Prompt Example: A vague request like "Help me prepare an interview" lacks clarity and detail, making it difficult for AI to provide useful responses.

• Structured Prompt Example: Clearly divided sections with explicit purposes enhance understanding and result in superior outputs from GPT models.

By following these guidelines on verbosity types and prompt structuring techniques, users can significantly improve their interactions with AI models like GPT-5.

Bucle de la Perfección: Mejora Continua con GPT

Introducción al Bucle de la Perfección

• El bucle de la perfección es un método avanzado y útil para tareas complejas, permitiendo aprovechar al máximo las capacidades de GPT-5.

• A diferencia del uso lineal tradicional (prompts y respuestas), este enfoque implica que el modelo cree una rúbrica de calidad antes de generar contenido.

Proceso del Bucle de la Perfección

• Se le pide a GPT que evalúe su propio trabajo utilizando la rúbrica creada, iterando hasta alcanzar la máxima puntuación en su percepción de excelencia.

• Este método es ideal para tareas creativas y complejas, como informes o estrategias detalladas, comenzando desde cero.

Ejemplos Prácticos

• Un ejemplo incluye pedir a GPT que desarrolle una rúbrica interna para un informe de análisis de mercado y luego refinarlo hasta obtener las mejores calificaciones.

• Otro ejemplo es crear una rúbrica para un plan de pruebas sobre nueva infraestructura, iterando hasta lograr un 10/10 en su evaluación.

Aplicación del Método

• Al aplicar el bucle con ejemplos prácticos, se observa cómo GPT itera y mejora sus resultados basándose en criterios establecidos previamente.

• La combinación de diferentes técnicas (como el metabromptim y el bucle de perfección) puede llevar a resultados aún más efectivos; el poder reside en la sinergia entre métodos.