Instruction-tuned LLM
Summary
Brief history
Instruction-tuned LLMs grew out of pretrained language models that could generate fluent text but did not reliably follow requests. InstructGPT later combined demonstrations with human-feedback training, helping establish the modern assistant model[1].
How it works
A transformer is first pretrained to predict the next token, then fine-tuned on instruction-and-response examples. Many systems also learn from ranked outputs to better align behavior with human intent[2]. At runtime, the model generates its answer token by token from the prompt and conversation context[2].
References
- “Language Models are Few-Shot Learners”. arxiv.org.
- “Training language models to follow instructions with human feedback”. arxiv.org.
Branch Outline
No Branch Outline is available for this thought.