Prompt Waywardness: The Curious Case of Discretized Interpretation of Continuous Prompts

Khashabi, Daniel; Lyu, Xinxi; Min, Sewon; Qin, Lianhui; Richardson, Kyle; Welleck, Sean; Hajishirzi, Hannaneh; Khot, Tushar; Sabharwal, Ashish; Singh, Sameer; Choi, Yejin

doi:10.18653/v1/2022.naacl-main.266

Cited by 11 publications

(9 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Prompt engineering, the templatized design of such prompts, can improve performance on the downstream task [70], e.g., code generation. According to Liu et al [70], approaches for prompting include appending relevant examples [41], appending a table schema [110], generating mutations of the prompt [64], summarizing complicated prompts [52], adding explanations [54], or various manipulations in the embedding space of the model [1,46,114]. Prompting guidelines for Codex 1 include specifying the language (e.g.…”

Section: Natural Language Programmingmentioning

confidence: 99%

“What It Wants Me To Say”: Bridging the Abstraction Gap Between End-User Programmers and Code-Generating Large Language Models

Liu¹,

Sarkar²,

Negreanu³

et al. 2023

Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems

View full text Add to dashboard Cite

Code-generating large language models map natural language to code. However, only a small portion of the infinite space of naturalistic utterances is effective at guiding code generation. For non-expert end-user programmers, learning this is the challenge of abstraction matching. We examine this challenge in the specific context of data analysis in spreadsheets, in a system that maps the user's natural language query to Python code using the Codex generator, executes the code, and shows the result. We propose grounded abstraction matching, which bridges the abstraction gap by translating the code back into a systematic and predictable naturalistic utterance. In a between-subjects, think-aloud study (n=24), we compare grounded abstraction matching to an ungrounded alternative based on previously established query framing principles. We find that the grounded approach improves end-users' understanding of the scope and capabilities of the code-generating model, and the kind of language needed to use it effectively. CCS CONCEPTS• Human-centered computing → Natural language interfaces; Interactive systems and tools; Empirical studies in HCI .

show abstract

Section: Natural Language Programmingmentioning

confidence: 99%

“What It Wants Me To Say”: Bridging the Abstraction Gap Between End-User Programmers and Code-Generating Large Language Models

Liu¹,

Sarkar²,

Negreanu³

et al. 2023

Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems

View full text Add to dashboard Cite

show abstract

“…These benchmarks focus on a single language and single dataset: the 2023) also release attribution annotations for 1.4M summaries, but the labels are machine-generated rather than human-annotated. GENIE (Khashabi et al, 2022) released 17K human evaluations across 5 tasks that includes one English summarization task (XSum).…”

Section: Related Workmentioning

confidence: 99%

SEAHORSE: A Multilingual, Multifaceted Dataset for Summarization Evaluation

Clark,

Rijhwani,

Gehrmann

et al. 2023

Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing

View full text Add to dashboard Cite

Reliable automatic evaluation of summarization systems is challenging due to the multifaceted and subjective nature of the task. This is especially the case for languages other than English, where human evaluations are scarce. In this work, we introduce SEAHORSE, a dataset for multilingual, multifaceted summarization evaluation. SEAHORSE consists of 96K summaries with human ratings along 6 dimensions of text quality: comprehensibility, repetition, grammar, attribution, main ideas, and conciseness. SEAHORSE covers 6 languages, 9 systems (including the reference text), and 4 summarization datasets. As a result of its size and scope, SEAHORSE can serve both as a benchmark to evaluate learnt metrics, as well as a large-scale resource for training such metrics. We show that metrics trained with SEAHORSE achieve strong performance on two out-of-domain meta-evaluation benchmarks: TRUE (Honovich et al., 2022) and mFACE . We make the SEAHORSE dataset and metrics publicly available for future research on multilingual and multifaceted summarization evaluation. 1

show abstract

“…On the one hand, it brings higher degrees of freedom for prompts, which leads to better results. On the other hand, those prompts become completely unknown to humans (Khashabi et al, 2022).…”

Section: Related Workmentioning

confidence: 99%

ReadPrompt: A Readable Prompting Method for Reliable Knowledge Probing

Wang,

Ye,

Wang

et al. 2023

Findings of the Association for Computational Linguistics: EMNLP 2023

View full text Add to dashboard Cite

Knowledge probing is a task to assess the knowledge encoded within pre-trained language models (PLMs) by having the PLM complete prompts such as "Italy is located in ____,". The model's prediction precision serves as a lower bound for the amount of knowledge it contains. Subsequent works explore training a series of vectors as prompts to guide PLMs towards more accurate predictions. However, these methods compromise the readability of the prompts. We cannot directly understand these prompts from their literal meaning, making it difficult to verify whether they are correct. Consequently, the credibility of probing results derived from these prompts is diminished. To address the issue, we propose a novel method called READPROMPT, which aims to identify meaningful sentences to serve as prompts. Experiments show that READ-PROMPT achieves state-of-the-art performance on the current knowledge probing benchmark. Moreover, since the prompt is readable, we discovered a misalignment between constructed prompts and knowledge, which is also present in current prompting methods verified by an attack experiment. We claim that the probing results obtained from the current prompting methods are unreliable and tend to overstate PLM's actual knowledge. 1

show abstract

Prompt Waywardness: The Curious Case of Discretized Interpretation of Continuous Prompts

Cited by 11 publications

References 20 publications

“What It Wants Me To Say”: Bridging the Abstraction Gap Between End-User Programmers and Code-Generating Large Language Models

“What It Wants Me To Say”: Bridging the Abstraction Gap Between End-User Programmers and Code-Generating Large Language Models

SEAHORSE: A Multilingual, Multifaceted Dataset for Summarization Evaluation

ReadPrompt: A Readable Prompting Method for Reliable Knowledge Probing

Contact Info

Product

Resources

About