Wiki Home

Zero-shot prompting

Definition

Zero-shot prompting asks a pretrained language model to perform a task from an instruction or partial input alone, with no labeled examples in the prompt and no weight updates. It exploits the finding that a model trained only to predict the next token acquires latent abilities a well-formed prompt can surface (Brown et al., 2020). Few-shot prompting differs by placing a handful of worked input-output examples in the context window before the query; both leave the weights untouched, which separates them from fine-tuning, where the model is further trained on task-specific data. Because the model continues the most likely text, the phrasing of the prompt strongly affects whether the continuation is what the user intended. For code, a zero-shot prompt is the surrounding source and comments with no example fixes attached.

Papers that use this concept

• Examining Zero-Shot Vulnerability Repair with Large Language Models — its central setting: repair vulnerabilities by prompting off-the-shelf models, with no fine-tuning on bug-fix data.

See also

• Language model pretraining
• Large language models for code

References

• Brown, T.B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. “Language Models are Few-Shot Learners.” arXiv:2005.14165, 2020.