Respuestas en lenguaje claro a las dudas de IA que se repiten. Se actualizan a medida que avanza el campo.
Un agente de IA es un modelo de lenguaje al que se le han dado herramientas que puede invocar, un objetivo que perseguir y permiso para dar varios pasos sin consultar a una persona entre uno y otro. Un chatbot responde y se detiene; un agente sigue actuando hasta que considera cumplido el objetivo o agota su presupuesto.
A context window is the maximum amount of text, measured in tokens, that a model can consider in a single request. It holds the system instructions, the conversation so far, any documents you paste in, and the answer being generated. When the total exceeds the limit, something has to be dropped or summarised.
Retrieval-augmented generation is a pattern where the system searches your own documents for passages relevant to a question, puts those passages into the model's prompt, and asks the model to answer using them. The model's weights never change; the knowledge arrives as context at request time.
Open weights means the trained model file can be downloaded and run yourself, under whatever licence the publisher chose. Open source is a stricter legal standard requiring freedom to use, study, modify and redistribute without restrictions on field of use. Many widely used models are open weights but not open source.
AI accelerators draw far more power per rack than traditional servers, and that power has to be delivered, cooled and paid for continuously. Training a large model is a one-off spike; serving it to millions of users is a permanent load, and inference is what dominates energy use over a deployed model's life.
Casi todas las API de IA facturan por millón de tokens, con precios separados para entrada y salida. La salida suele costar varias veces más que la entrada. La entrada cacheada, el procesamiento por lotes y los modelos más pequeños reducen la factura de forma sustancial, y el coste de una conversación crece con el historial porque la mayoría de las API reenvían el hilo completo en cada turno.