designed for understanding and generating human language.
based on deep learning architectures
such as Transformers
trained on massive amounts of text data from various sources
acquire a deep understanding of the nuances (细微差距) and complexities of language.
have the ability to achieve state-of-the-art (最先进) performance in multiple Natural Language Processing (NLP) tasks
such as machine translation, sentiment (情绪) analysis, summarization, and more.
They can also generate coherent (连贯一致的) and contextually relevant text based on given input, making them highly useful for applications like chatbots, question-answering systems, and content generation.
Types
Base LLMs
designed to predict the next word based on the training data.
not designed to answer questions
Instruction tuned (指令调优) LLMs
Instruction Tuned LLMs = Base LLMs + Further Tuning + RLHF
Further Tuning: trained using a large dataset covering sample “Instructions” and how the model should perform as a result of those instructions.
Reinforcement Learning (强化学习) with Human Feedback (RLHF)
