REVIEW 9 cited by
Demystifying Prompts in Language Models via Perplexity Estimation
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Language models can be prompted to perform a wide variety of zero- and few-shot learning problems. However, performance varies significantly with the choice of prompt, and we do not yet understand why this happens or how to pick the best prompts. In this work, we analyze the factors that contribute to this variance and establish a new empirical hypothesis: the performance of a prompt is coupled with the extent to which the model is familiar with the language it contains. Over a wide range of tasks, we show that the lower the perplexity of the prompt is, the better the prompt is able to perform the task. As a result, we devise a method for creating prompts: (1) automatically extend a small seed set of manually written prompts by paraphrasing using GPT3 and backtranslation and (2) choose the lowest perplexity prompts to get significant gains in performance.
Forward citations
Cited by 9 Pith papers
-
DisarmRAG: Stealthy Retriever-Centric Poisoning to Disable Self-Correction in Retrieval-Augmented Generation (Extended Version)
DisarmRAG compromises the retriever to inject anti-self-correction instructions, achieving over 90% attack success across six LLMs while evading basic detection.
-
When Compression Becomes an Attack Surface: Black-Box Attacks on Prompt-Compressed LLM Agents
The paper claims prompt compression is a new attack surface, but the abstract's COMA attack never appears in the body and the body's SoftCom requires white-box access.
-
How Important is `Perfect' English for Machine Translation Prompts?
For LLM machine translation, prompt choice affects output quality more than realistic user errors, with spelling errors hurting most and phrase-level errors often harmless.
-
CoTSRF: Utilize Chain of Thought as Stealthy and Robust Fingerprint of Large Language Models
CoTSRF fingerprints a source LLM by training a contrastive encoder on chain-of-thought responses, then flags suspect APIs whose reasoning-style feature distances are too close to the source's distribution.
-
Practical, Generalizable and Robust Backdoor Attacks on Text-to-Image Diffusion Models
The abstract claims that 10 poisoned samples can backdoor multiple text-to-image models with over 90% attack success and resistance to defenses, but the supplied body is a different paper.
-
DICE: Dynamic In-Context Example Selection in LLM Agents via Efficient Knowledge Transfer
DICE dynamically retrieves the most relevant in-context demonstrations at each agent step, and in this preprint it raises exact-match and success-rate scores on HotpotQA, ALFWorld, and Webshop across ReAct, Reflexion,...
-
Unveiling Effective In-Context Configurations for Image Captioning: An External & Internal Analysis
For Flamingo-style models, increasing the number of in-context examples improves language coherence but degrades visual-text alignment, and similarity-based image retrieval inflates CIDEr scores by encouraging caption...
-
Low-Perplexity LLM-Generated Sequences and Where To Find Them
Only about 40% of low-perplexity 6-token spans generated by Pythia-6.9B can be exactly matched to The Pile, and the authors categorize matched and unmatched spans into four classes.
-
Large Language Models in Cybersecurity: Applications, Vulnerabilities, and Defense Techniques
A survey that maps LLM applications, vulnerabilities, and defenses across eight cybersecurity domains, but with significant citation and rigor problems.
Discussion (0). Sign in to comment.