AI-ordlista
Den kompletta ordlistan över AI
Instruction Tuning
Process of fine-tuning a pre-trained model on instruction-response pairs to improve its ability to follow instructions. This approach optimizes the model for zero-shot and few-shot learning.
Meta-Prompting
Advanced strategy where the prompt itself is generated or optimized by another model or automated process. This approach allows for dynamic adaptation of prompts to the specificities of each task.
Zero-Shot Transfer
Ability of a model to apply knowledge learned from one task or domain to a completely different task without specific examples. This skill is crucial for large-scale generalization.
Few-Shot Adaptation
Process by which a model quickly adjusts its behavior from a minimal number of examples for a new task. Adaptation occurs at the activation level without modifying the network weights.
Prompt Calibration
Fine-tuning technique for prompts to align the model's probability distributions with specific task expectations. Calibration improves the reliability and consistency of predictions.
Contextual Prompting
Prompting approach that dynamically integrates relevant context into the prompt to guide the model toward more accurate responses. This method adapts the prompt based on available information.
Multi-Shot Learning
Variant of few-shot learning using a moderate number of examples (typically 5-20) to optimize in-context learning. This approach balances efficiency and performance on complex tasks.
Adaptive Prompting
Prompting system that automatically adjusts prompts based on the model's previous responses and performance metrics. This dynamic adaptation optimizes real-time interaction.
Prompt Ensemble
Technique using multiple different prompts for the same task and combining their results to improve robustness and accuracy. The prompt ensemble leverages various perspectives on the problem.