Eliciting Knowledge from Language Models Using Automatically Generated Prompts

By Taylor Shin, Yasaman Razeghi, Robert L Logan IV, Eric Wallace, Sameer Singh. 2020

559 citations recorded by the source corpus.

The remarkable success of pretrained language models has motivated the study of what kinds of knowledge these models learn during pretraining. Reformulating tasks as fill-in-the-blanks problems (e.g., cloze tests) is a natural approach for gauging such knowledge, however, its usage is limited by the manual effort and guesswork required to write suitable prompts. To address this, we develop AutoPrompt, an automated method to create prompts for a diverse set of tasks, based on a gradient-guided search. Using AutoPrompt, we show that masked language models (MLMs) have an inherent capability to perform sentiment analysis and natural language inference without additional parameters or finetuning, sometimes achieving performance on par with recent state-of-the-art supervised models. We also show that our prompts elicit more accurate factual knowledge from MLMs than the manually created prompts on the LAMA benchmark, and that MLMs can be used as relation extractors more effectively than supervised relation extraction models. These results demonstrate that automatically generated prompts are a viable parameter-free alternative to existing probing methods, and as pretrained LMs become more sophisticated and capable, potentially a replacement for finetuning.

Subfields: Artificial Intelligence, Large Language Models, Prompting & Reasoning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Large Language Models

In Prompting & Reasoning

Citations in the corpus

AI Chains: Transparent and Controllable Human-AI Interaction by Chaining Large Language Model Prompts (2021, background): “However, the mined prompts tend to be less human-readable [61] and therefore less compatible with human-AI interaction.”
Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing (2021, methodology): “[125] learn a logistic classifier using the contextualized representation of the [Z] token as input.”
Calibrate Before Use: Improving Few-Shot Performance of Language Models (2021, methodology): “Asking LMs to complete natural language prompts is also used as a method to “probe” LMs, e.g., analyzing their factual (Petroni et al., 2019; Jiang et al., 2020b; Shin et al., 2020) or commonsense knowledge (Bosselut et al., 2019).”
Learning to Prompt for Vision-Language Models (2021, methodology): “Though the performance is excellent, the results of CoOp are relatively diﬃcult to interpret, like other continuous prompt learning methods in NLP.”
Unifying Vision-and-Language Tasks via Text Generation (2021, methodology): “Dataset-speciﬁc preﬁxes: As shown in recent works (Gao et al., 2020; Shin et al., 2020; Li & Liang, 2021; Radford et al., 2021), different text prompts could result in different ﬁnetuning results.”
CLIP-Adapter: Better Vision-Language Models with Feature Adapters (2021, neutral): “As the earliest attempt, one type of them focus on tuning pretrained language or vision-language models with natural language discrete prompts [1, 17, 61, 65, 68, 69] or prompt engineering by mining or generating proper natural language discrete prompts [17, 29, 53].”
Neural Data Augmentation via Example Extrapolation (2021, background): “Therefore, GPT-3 users must be careful to format examples in a way that resembles “natural” text encountered during pre-training – such “format engineering” can greatly affect performance (Shin et al., 2020; Schick and Schütze, 2020).”
An Explanation of In-context Learning as Implicit Bayesian Inference (2021, positive): “Finetun-ing LMs on text with a prompting format improves its zero-shot performance (Wei et al., 2021b; Sanh et al., 2021) and optimizing prompt templates improves few-shot ﬁnetuning (Jiang et al., 2020; Schick & Sch¨utze, 2021; Shin et al., 2020; Gao et al., 2021).”
Reframing Instructional Prompts to GPTk’s Language (2021, background): “(3) Existing proposals to discover prompts (Shin et al., 2020; Jiang et al., 2020) lead to results that are model-specific, i.”
PTR: Prompt Tuning with Rules for Text Classification (2021, background): “Shin et al. (2020) further explore gradient-guided search to automatically generate both templates and label words.”
Prompt-free and Efficient Few-shot Learning with Language Models (2022, background): “…setting, while being inefficient and not as effective compared to adapters (Ma-habadi et al., 2021a). b) Optimizing patterns in a discrete space (Shin et al., 2020; Jiang et al., 2020; Gao et al., 2021) has the problem that such methods are computationally costly. c) Automatically finding…”
Language in a Bottle: Language Model Guided Concept Bottlenecks for Interpretable Image Classification (2022, methodology): “These focus on searching over text prompts to improve classification performance, and resemble earlier techniques in LLM prompt tuning [15, 53, 56].”
Security and Privacy Challenges of Large Language Models: A Survey (2024, unknown): “While several studies focused on manual or experimental prompt injection techniques, Shin et al. introduced AutoPrompt, an automated approach to prompt generation for diverse tasks employing a gradient-guided search [51] strategy to obtain an efficient prompt template.”
CPT: Colorful Prompt Tuning for Pre-trained Vision-Language Models (2021, unknown): “To facilitate prompt engineering, Shin et al. (2020) pro-pose to automatically generate prompt templates via gradient-based search.”