We need to understand that not all models are created equal. The main difference between general models and instructional models.
General models
Ejemplo:
Pregunta: “Explain how a neural network works”
Response from a general model: can give an extensive explanation, include unnecessary concepts or jump topics.
Instruct Models
Ejemplo:
Pregunta: “Explain how a neural network works step by step for beginners”
Response from an instructional model: clear explanation, structured in steps, with simple examples and audience-centered.
Summary Comparison:
| Characteristic | General Model | Instructed Model |
|---|---|---|
| Objective | Predict fluid text | Follow instructions and provide precise answers |
| Best use case | Creative writing, free generation | Chatbots, assistants, technical tasks |
| Relevance | May diverge | High, responds to the request |
| Context | Limited | Maintains and understands context |
| Example | “Coherent text” | “Specific response to your question” |
If your goal is to generate free or creative text, a general model may be sufficient.
If you need precise, coherent and instruction-adapted responses, especially for technical chatbots or AI agents, instructed models are the most effective option.
In practice, tools such as Llama 3 Instruct, Mistral Mixtral or Alpaca are ideal for environments where instruction understanding is critical.
What model to use?
General models
Designed for free text generation, completion, creativity, and tasks where you don’t need precise instruction following:
Models of instructions (Instruct Models)
Fine-tuned to understand and follow instructions, ideal for chatbots, assistants or AI agents:
| Type of model | ||
|---|---|---|
| General | Llama 3, BLOOM, GPT-NeoX, MPT-7B | Text generation, creative writing, summaries, translations |
| Instruct | Llama 3 Instruct, Mistral Mixtral, Alpaca/Vicuna, Phi-3 Mini, Gemma 7B | Chatbots, assistants, AI agents, technical QA, guided responses |
