An official website of the United States government
In this paper, we develop a method for extracting information from Large Language Models (LLMs) with associated confidence estimates. We propose that effective confidence models may be designed using a large number of uncertainty measures (i.e., variables that are only weakly predictive of - but positively correlated with - information correctness) as inputs. We trained a confidence model that uses 20 handcrafted uncertainty measures to predict GPT-4’s ability to reproduce species occurrence data from iDigBio and found that, if we only consider occurrence claims that are placed in the top 30% of confidence estimates, we can increase prediction accuracy from 57% to 88% for species absence predictions and from 77% to 86% for species presence predictions. Using the same confidence model, we used GPT- 4 to extract new data that extrapolates beyond the occurrence records in iDigBio and used the results to visualize geographic distributions for four individual species. More generally, this represents a novel use case for LLMs in generating credible pseudo data for applications in which high-quality curated data are unavailable or inaccessible.
more »
« less
Tracing the Evolution of Information Transparency for OpenAI’s GPT Models through a Biographical Approach
Information transparency, the open disclosure of information about models, is crucial for proactively evaluating the potential societal harm of large language models (LLMs) and developing effective risk mitigation measures. Adapting the biographies of artifacts and practices (BOAP) method from science and technology studies, this study analyzes the evolution of information transparency within OpenAI’s Generative Pre-trained Transformers (GPT) model reports and usage policies from its inception in 2018 to GPT-4, one of today’s most capable LLMs. To assess the breadth and depth of transparency practices, we develop a 9-dimensional, 3-level analytical framework to evaluate the comprehensiveness and accessibility of information disclosed to various stakeholders. Findings suggest that while model limitations and downstream usages are increasingly clarified, model development processes have become more opaque. Transparency remains minimal in certain aspects, such as model explainability and real-world evidence of LLM impacts, and the discussions on safety measures such as technical interventions and regulation pipelines lack in-depth details. The findings emphasize the need for enhanced transparency to foster accountability and ensure responsible technological innovations.
more »
« less
- Award ID(s):
- 2220772
- PAR ID:
- 10560251
- Editor(s):
- Das, Sanmay; Green, Brian Patrick; Varshney, Kush; Ganapini, Marianna; Renda, Andrea
- Publisher / Repository:
- AAAI
- Date Published:
- Journal Name:
- Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society
- Volume:
- 7
- ISSN:
- 3065-8365
- Page Range / eLocation ID:
- 1684 to 1695
- Subject(s) / Keyword(s):
- information transparency LLMs OpenAI STS
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Safety is critical to the usage of large language models (LLMs). Multiple techniques such as data filtering and supervised fine tuning have been developed to strengthen LLM safety. However, currently known techniques presume that corpora used for safety alignment of LLMs are solely interpreted by semantics. This assumption, however, does not hold in real-world applications, which leads to severe vulnerabilities in LLMs. For example, users of forums often use ASCII art, a form of text-based art, to convey image information. In this paper, we propose a novel ASCII art-based jailbreak attack and introduce a comprehensive benchmark Vision-in-Text Challenge (VITC) to evaluate the capabilities of LLMs in recognizing prompts that cannot be solely interpreted by semantics. We show that five SOTA LLMs (GPT-3.5, GPT-4, Gemini, Claude, and Llama2) struggle to recognize prompts provided in the form of ASCII art. Based on this observation, we develop the jailbreak attack ArtPrompt, which leverages the poor performance of LLMs in recognizing ASCII art to bypass safety measures and elicit undesired behaviors from LLMs. ArtPrompt only requires black-box access to the victim LLMs, making it a practical attack. We evaluate ArtPrompt on five SOTA LLMs, and show that ArtPrompt can effectively and efficiently induce undesired behaviors from all five LLMs. Our code is available at https: //github.com/uw-nsl/ArtPrompt.more » « less
-
Large Language Models (LLMs) are often augmented with external contexts, such as those used in retrieval-augmented generation (RAG). However, these contexts can be inaccurate or intentionally misleading, leading to conflicts with the model’s internal knowledge. We argue that robust LLMs should demonstrate situated faithfulness, dynamically calibrating their trust in external information based on their confidence in the internal knowledge and the external context to resolve knowledge conflicts. To benchmark this capability, we evaluate LLMs across several QA datasets, including a newly created dataset featuring in-the-wild incorrect contexts sourced from Reddit posts. We show that when provided with both correct and incorrect contexts, both open-source and proprietary models tend to overly rely on external information, regardless of its factual accuracy. To enhance situated faithfulness, we propose two approaches: Self-Guided Confidence Reasoning (SCR) and Rule-Based Confidence Reasoning (RCR). SCR enables models to self-access the confidence of external information relative to their own internal knowledge to produce the most accurate answer. RCR, in contrast, extracts explicit confidence signals from the LLM and determines the final answer using predefined rules. Our results show that for LLMs with strong reasoning capabilities, such as GPT-4o and GPT-4o mini, SCR outperforms RCR, achieving improvements of up to 24.2% over a direct input augmentation baseline. Conversely, for a smaller model like Llama-3-8B, RCR outperforms SCR. Fine-tuning SCR with our proposed Confidence Reasoning Direct Preference Optimization (CR-DPO) method improves performance on both seen and unseen datasets, yielding an average improvement of 8.9% on Llama-3-8B. In addition to quantitative results, we offer insights into the relative strengths of SCR and RCR. Our findings highlight promising avenues for improving situated faithfulness in LLMs.more » « less
-
Bonial, Claire; Bonn, Julia; Hwang, Jena D (Ed.)We evaluate the ability of large language models (LLMs) to provide PropBank semantic role label annotations across different realizations of the same verbs in transitive, intransitive, and middle voice constructions. In order to assess the meta-linguistic capabilities of LLMs as well as their ability to glean such capabilities through in-context learning, we evaluate the models in a zero-shot setting, in a setting where it is given three examples of another verb used in transitive, intransitive, and middle voice constructions, and finally in a setting where it is given the examples as well as the correct sense and roleset information. We find that zero-shot knowledge of PropBank annotation is almost nonexistent. The largest model evaluated, GPT-4, achieves the best performance in the setting where it is given both examples and the correct roleset in the prompt, demonstrating that larger models can ascertain some meta-linguistic capabilities through in-context learning. However, even in this setting, which is simpler than the task of a human in PropBank annotation, the model achieves only 48% accuracy in marking numbered arguments correctly. To ensure transparency and reproducibility, we publicly release our dataset and model responses.more » « less
-
Advances in large language models (LLMs) have empowered a variety of applications. However, there is still a significant gap in research when it comes to understanding and enhancing the capabilities of LLMs in the field of mental health. In this work, we present a comprehensive evaluation of multiple LLMs on various mental health prediction tasks via online text data, including Alpaca, Alpaca-LoRA, FLAN-T5, GPT-3.5, and GPT-4. We conduct a broad range of experiments, covering zero-shot prompting, few-shot prompting, and instruction fine-tuning. The results indicate a promising yet limited performance of LLMs with zero-shot and few-shot prompt designs for mental health tasks. More importantly, our experiments show that instruction finetuning can significantly boost the performance of LLMs for all tasks simultaneously. Our best-finetuned models, Mental-Alpaca and Mental-FLAN-T5, outperform the best prompt design of GPT-3.5 (25 and 15 times bigger) by 10.9% on balanced accuracy and the best of GPT-4 (250 and 150 times bigger) by 4.8%. They further perform on par with the state-of-the-art task-specific language model. We also conduct an exploratory case study on LLMs' capability on mental health reasoning tasks, illustrating the promising capability of certain models such as GPT-4. We summarize our findings into a set of action guidelines for potential methods to enhance LLMs' capability for mental health tasks. Meanwhile, we also emphasize the important limitations before achieving deployability in real-world mental health settings, such as known racial and gender bias. We highlight the important ethical risks accompanying this line of research.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1609/aies.v7i1.31757
