Free Information Center
Jailbreak in large language models refers to techniques used to bypass or override the built-in restrictions and safety mechanisms of AI systems to produce outputs that are otherwise restricted. These methods raise ethical, security, and safety concerns within the AI community.
QLoRA (quantized LoRA) is a machine learning technique that combines quantization and Low-Rank Adaptation to efficiently fine-tune large language models with reduced computational resources while maintaining performance.
Zero-shot prompting is a technique in natural language processing where a model is given instructions or queries without prior examples, enabling it to perform tasks it was not explicitly trained on. This approach leverages the generalization capabilities of large language models to interpret and respond to new prompts directly.
LMQL (Language Model Query Language) is a programming language designed to facilitate precise interaction and querying of large language models. It enables developers to write structured queries that integrate with language models to control output generation effectively.
Tree-of-thought prompting is a technique used in artificial intelligence to improve reasoning and decision-making by structuring the problem-solving process as a tree of interconnected thoughts or steps. It enhances large language models’ ability to explore multiple reasoning paths and generate more accurate or creative outputs.
Instruction tuning is a technique in machine learning where models are fine-tuned on datasets containing task instructions, enhancing their ability to follow diverse prompts and perform various tasks. This approach improves model generalization and adaptability across multiple applications.
Prompt engineering is the practice of designing and refining input prompts for artificial intelligence models, especially large language models, to achieve desired outputs. It involves crafting queries or instructions that guide AI systems to generate more accurate, relevant, or contextually appropriate responses.
BIG-bench is a large-scale benchmark designed to evaluate the capabilities of language models across diverse and challenging tasks. It aims to provide a comprehensive assessment of model performance beyond conventional benchmarks.