Building a Natural Language Interface for FHIR Clinical Terminology Server.

While Fast Healthcare Interoperability Resources (FHIR) clinical terminology server enables quick and easy search and retrieval of coded medical data, it still has some drawbacks. When searching, any typographical errors, variations in word forms, or deviations in word sequence might lead to incorrect search outcomes. For retrieval, queries to the server must strictly follow the FHIR application programming interface format, which requires users to know the syntax and remember the attribute codes they wish to retrieve. To improve its functionalities, a natural language interface was built, that harnesses the capabilities of two preeminent large language models, along with other cutting-edge technologies such as speech-to-text conversion, vector semantic searching, and conversational artificial intelligence. Preliminary evaluation shows promising results in building a natural language interface for the FHIR clinical terminology system.

Speech Emotion Recognition Using openSMILE and GPT 3.5 Transformer.

In recent years, artificial intelligence, and machine learning (ML) models have advanced significantly, offering transformative solutions across diverse sectors. Emotion recognition in speech has particularly benefited from ML techniques, revolutionizing its accuracy and applicability. This article proposes a method for emotion detection in Romanian speech analysis by combining two distinct approaches: semantic analysis using GPT Transformer and acoustic analysis using openSMILE. The results showed an accuracy of 74% and a precision of almost 82%. Several system limitations were observed due to the limited and low-quality dataset. However, it also opened a new horizon in our research by analyzing emotions to identify mental health disorders.

Human-Comparable Sensitivity of Large Language Models in Identifying Eligible Studies Through Title and Abstract Screening: 3-Layer Strategy Using GPT-3.5 and GPT-4 for Systematic Reviews.

BACKGROUND: The screening process for systematic reviews is resource-intensive. Although previous machine learning solutions have reported reductions in workload, they risked excluding relevant papers. OBJECTIVE: We evaluated the performance of a 3-layer screening method using GPT-3.5 and GPT-4 to streamline the title and abstract-screening process for systematic reviews. Our goal is to develop a screening method that maximizes sensitivity for identifying relevant records. METHODS: We conducted screenings on 2 of our previous systematic reviews related to the treatment of bipolar disorder, with 1381 records from the first review and 3146 from the second. Screenings were conducted using GPT-3.5 (gpt-3.5-turbo-0125) and GPT-4 (gpt-4-0125-preview) across three layers: (1) research design, (2) target patients, and (3) interventions and controls. The 3-layer screening was conducted using prompts tailored to each study. During this process, information extraction according to each study's inclusion criteria and optimization for screening were carried out using a GPT-4-based flow without manual adjustments. Records were evaluated at each layer, and those meeting the inclusion criteria at all layers were subsequently judged as included. RESULTS: On each layer, both GPT-3.5 and GPT-4 were able to process about 110 records per minute, and the total time required for screening the first and second studies was approximately 1 hour and 2 hours, respectively. In the first study, the sensitivities/specificities of the GPT-3.5 and GPT-4 were 0.900/0.709 and 0.806/0.996, respectively. Both screenings by GPT-3.5 and GPT-4 judged all 6 records used for the meta-analysis as included. In the second study, the sensitivities/specificities of the GPT-3.5 and GPT-4 were 0.958/0.116 and 0.875/0.855, respectively. The sensitivities for the relevant records align with those of human evaluators: 0.867-1.000 for the first study and 0.776-0.979 for the second study. Both screenings by GPT-3.5 and GPT-4 judged all 9 records used for the meta-analysis as included. After accounting for justifiably excluded records by GPT-4, the sensitivities/specificities of the GPT-4 screening were 0.962/0.996 in the first study and 0.943/0.855 in the second study. Further investigation indicated that the cases incorrectly excluded by GPT-3.5 were due to a lack of domain knowledge, while the cases incorrectly excluded by GPT-4 were due to misinterpretations of the inclusion criteria. CONCLUSIONS: Our 3-layer screening method with GPT-4 demonstrated acceptable level of sensitivity and specificity that supports its practical application in systematic review screenings. Future research should aim to generalize this approach and explore its effectiveness in diverse settings, both medical and nonmedical, to fully establish its use and operational feasibility.

Text summarization for pharmaceutical sciences using hierarchical clustering with a weighted evaluation methodology.

In the pharmaceutical industry, there is an abundance of regulatory documents used to understand the current regulatory landscape and proactively make project decisions. Due to the size of these documents, it is helpful for project teams to have informative summaries. We propose a novel solution, MedicoVerse, to summarize such documents using advanced machine learning techniques. MedicoVerse uses a multi-stage approach, combining word embeddings using the SapBERT model on regulatory documents. These embeddings are put through a critical hierarchical agglomerative clustering step, and the clusters are organized through a custom data structure. Each cluster is summarized using the bart-large-cnn-samsum model, and each summary is merged to create a comprehensive summary of the original document. We compare MedicoVerse results with established models T5, Google Pegasus, Facebook BART, and large language models such as Mixtral 8 × 7b instruct, GPT 3.5, and Llama-2-70b by introducing a scoring system that considers four factors: ROUGE score, BERTScore, business entities and the Flesch Reading Ease. Our results show that MedicoVerse outperforms the compared models, thus producing informative summaries of large regulatory documents.

Evaluating prompt engineering on GPT-3.5's performance in USMLE-style medical calculations and clinical scenarios generated by GPT-4.

This study was designed to assess how different prompt engineering techniques, specifically direct prompts, Chain of Thought (CoT), and a modified CoT approach, influence the ability of GPT-3.5 to answer clinical and calculation-based medical questions, particularly those styled like the USMLE Step 1 exams. To achieve this, we analyzed the responses of GPT-3.5 to two distinct sets of questions: a batch of 1000 questions generated by GPT-4, and another set comprising 95 real USMLE Step 1 questions. These questions spanned a range of medical calculations and clinical scenarios across various fields and difficulty levels. Our analysis revealed that there were no significant differences in the accuracy of GPT-3.5's responses when using direct prompts, CoT, or modified CoT methods. For instance, in the USMLE sample, the success rates were 61.7% for direct prompts, 62.8% for CoT, and 57.4% for modified CoT, with a p-value of 0.734. Similar trends were observed in the responses to GPT-4 generated questions, both clinical and calculation-based, with p-values above 0.05 indicating no significant difference between the prompt types. The conclusion drawn from this study is that the use of CoT prompt engineering does not significantly alter GPT-3.5's effectiveness in handling medical calculations or clinical scenario questions styled like those in USMLE exams. This finding is crucial as it suggests that performance of ChatGPT remains consistent regardless of whether a CoT technique is used instead of direct prompts. This consistency could be instrumental in simplifying the integration of AI tools like ChatGPT into medical education, enabling healthcare professionals to utilize these tools with ease, without the necessity for complex prompt engineering.

A comprehensive dataset for Arabic word sense disambiguation.

This data paper introduces a comprehensive dataset tailored for word sense disambiguation tasks, explicitly focusing on a hundred polysemous words frequently employed in Modern Standard Arabic. The dataset encompasses a diverse set of senses for each word, ranging from 3 to 8, resulting in 367 unique senses. Each word sense is accompanied by contextual sentences comprising ten sentence examples that feature the polysemous word in various contexts. The data collection resulted in a dataset of 3670 samples. Significantly, the dataset is in Arabic, which is known for its rich morphology, complex syntax, and extensive polysemy. The data was meticulously collected from various web sources, spanning news, medicine, finance, and more domains. This inclusivity ensures the dataset's applicability across diverse fields, positioning it as a pivotal resource for Arabic Natural Language Processing (NLP) applications. The data collection timeframe spans from the first of April 2023 to the first of May 2023. The dataset provides comprehensive model learning by including all senses for a frequently used Arabic polysemous term, even rare senses that are infrequently used in real-world contexts, thereby mitigating biases. The dataset comprises synthetic sentences generated by GPT3.5-turbo, addressing instances where rare senses lack sufficient real-world data. The dataset collection process involved initial web scraping, followed by manual sorting to distinguish word senses, supplemented by thorough searches by a human expert to fill in missing contextual sentences. Finally, in instances where online data for rare word senses was lacking or insufficient, synthetic samples were generated. Beyond its primary utility in word sense disambiguation, this dataset holds considerable value for scientists and researchers across various domains, extending its relevance to sentiment analysis applications.

A Study of Biomedical Relation Extraction Using GPT Models.

Relation Extraction (RE) is a natural language processing (NLP) task for extracting semantic relations between biomedical entities. Recent developments in pre-trained large language models (LLM) motivated NLP researchers to use them for various NLP tasks. We investigated GPT-3.5-turbo and GPT-4 on extracting the relations from three standard datasets, EU-ADR, Gene Associations Database (GAD), and ChemProt. Unlike the existing approaches using datasets with masked entities, we used three versions for each dataset for our experiment: a version with masked entities, a second version with the original entities (unmasked), and a third version with abbreviations replaced with the original terms. We developed the prompts for various versions and used the chat completion model from GPT API. Our approach achieved a F1-score of 0.498 to 0.809 for GPT-3.5-turbo, and a highest F1-score of 0.84 for GPT-4. For certain experiments, the performance of GPT, BioBERT, and PubMedBERT are almost the same.

Using large language models for safety-related table summarization in clinical study reports.

Objectives: The generation of structured documents for clinical trials is a promising application of large language models (LLMs). We share opportunities, insights, and challenges from a competitive challenge that used LLMs for automating clinical trial documentation. Materials and Methods: As part of a challenge initiated by Pfizer (organizer), several teams (participant) created a pilot for generating summaries of safety tables for clinical study reports (CSRs). Our evaluation framework used automated metrics and expert reviews to assess the quality of AI-generated documents. Results: The comparative analysis revealed differences in performance across solutions, particularly in factual accuracy and lean writing. Most participants employed prompt engineering with generative pre-trained transformer (GPT) models. Discussion: We discuss areas for improvement, including better ingestion of tables, addition of context and fine-tuning. Conclusion: The challenge results demonstrate the potential of LLMs in automating table summarization in CSRs while also revealing the importance of human involvement and continued research to optimize this technology.

Evaluating the OpenAI's GPT-3.5 Turbo's performance in extracting information from scientific articles on diabetic retinopathy.

We aimed to compare the concordance of information extracted and the time taken between a large language model (OpenAI's GPT-3.5 Turbo via API) against conventional human extraction methods in retrieving information from scientific articles on diabetic retinopathy (DR). The extraction was done using GPT3.5 Turbo as of October 2023. OpenAI's GPT-3.5 Turbo significantly reduced the time taken for extraction. Concordance was highest at 100% for the extraction of the country of study, 64.7% for significant risk factors of DR, 47.1% for exclusion and inclusion criteria, and lastly 41.2% for odds ratio (OR) and 95% confidence interval (CI). The concordance levels seemed to indicate the complexity associated with each prompt. This suggests that OpenAI's GPT-3.5 Turbo may be adopted to extract simple information that is easily located in the text, leaving more complex information to be extracted by the researcher. It is crucial to note that the foundation model is constantly improving significantly with new versions being released quickly. Subsequent work can focus on retrieval-augmented generation (RAG), embedding, chunking PDF into useful sections, and prompting to improve the accuracy of extraction.

Distilling large language models for matching patients to clinical trials.

OBJECTIVE: The objective of this study is to systematically examine the efficacy of both proprietary (GPT-3.5, GPT-4) and open-source large language models (LLMs) (LLAMA 7B, 13B, 70B) in the context of matching patients to clinical trials in healthcare. MATERIALS AND METHODS: The study employs a multifaceted evaluation framework, incorporating extensive automated and human-centric assessments along with a detailed error analysis for each model, and assesses LLMs' capabilities in analyzing patient eligibility against clinical trial's inclusion and exclusion criteria. To improve the adaptability of open-source LLMs, a specialized synthetic dataset was created using GPT-4, facilitating effective fine-tuning under constrained data conditions. RESULTS: The findings indicate that open-source LLMs, when fine-tuned on this limited and synthetic dataset, achieve performance parity with their proprietary counterparts, such as GPT-3.5. DISCUSSION: This study highlights the recent success of LLMs in the high-stakes domain of healthcare, specifically in patient-trial matching. The research demonstrates the potential of open-source models to match the performance of proprietary models when fine-tuned appropriately, addressing challenges like cost, privacy, and reproducibility concerns associated with closed-source proprietary LLMs. CONCLUSION: The study underscores the opportunity for open-source LLMs in patient-trial matching. To encourage further research and applications in this field, the annotated evaluation dataset and the fine-tuned LLM, Trial-LLAMA, are released for public use.

Evaluation of Large Language Model Performance and Reliability for Citations and References in Scholarly Writing: Cross-Disciplinary Study.

BACKGROUND: Large language models (LLMs) have gained prominence since the release of ChatGPT in late 2022. OBJECTIVE: The aim of this study was to assess the accuracy of citations and references generated by ChatGPT (GPT-3.5) in two distinct academic domains: the natural sciences and humanities. METHODS: Two researchers independently prompted ChatGPT to write an introduction section for a manuscript and include citations; they then evaluated the accuracy of the citations and Digital Object Identifiers (DOIs). Results were compared between the two disciplines. RESULTS: Ten topics were included, including 5 in the natural sciences and 5 in the humanities. A total of 102 citations were generated, with 55 in the natural sciences and 47 in the humanities. Among these, 40 citations (72.7%) in the natural sciences and 36 citations (76.6%) in the humanities were confirmed to exist (P=.42). There were significant disparities found in DOI presence in the natural sciences (39/55, 70.9%) and the humanities (18/47, 38.3%), along with significant differences in accuracy between the two disciplines (18/55, 32.7% vs 4/47, 8.5%). DOI hallucination was more prevalent in the humanities (42/55, 89.4%). The Levenshtein distance was significantly higher in the humanities than in the natural sciences, reflecting the lower DOI accuracy. CONCLUSIONS: ChatGPT's performance in generating citations and references varies across disciplines. Differences in DOI standards and disciplinary nuances contribute to performance variations. Researchers should consider the strengths and limitations of artificial intelligence writing tools with respect to citation accuracy. The use of domain-specific models may enhance accuracy.

Performance Assessment of ChatGPT versus Bard in Detecting Alzheimer's Dementia.

Large language models (LLMs) find increasing applications in many fields. Here, three LLM chatbots (ChatGPT-3.5, ChatGPT-4, and Bard) are assessed in their current form, as publicly available, for their ability to recognize Alzheimer's dementia (AD) and Cognitively Normal (CN) individuals using textual input derived from spontaneous speech recordings. A zero-shot learning approach is used at two levels of independent queries, with the second query (chain-of-thought prompting) eliciting more detailed information than the first. Each LLM chatbot's performance is evaluated on the prediction generated in terms of accuracy, sensitivity, specificity, precision, and F1 score. LLM chatbots generated a three-class outcome ("AD", "CN", or "Unsure"). When positively identifying AD, Bard produced the highest true-positives (89% recall) and highest F1 score (71%), but tended to misidentify CN as AD, with high confidence (low "Unsure" rates); for positively identifying CN, GPT-4 resulted in the highest true-negatives at 56% and highest F1 score (62%), adopting a diplomatic stance (moderate "Unsure" rates). Overall, the three LLM chatbots can identify AD vs. CN, surpassing chance-levels, but do not currently satisfy the requirements for clinical application.

Large language models and bariatric surgery patient education: a comparative readability analysis of GPT-3.5, GPT-4, Bard, and online institutional resources.

BACKGROUND: The readability of online bariatric surgery patient education materials (PEMs) often surpasses the recommended 6th grade level. Large language models (LLMs), like ChatGPT and Bard, have the potential to revolutionize PEM delivery. We aimed to evaluate the readability of PEMs produced by U.S. medical institutions compared to LLMs, as well as the ability of LLMs to simplify their responses. METHODS: Responses to frequently asked questions (FAQs) related to bariatric surgery were gathered from top-ranked health institutions. FAQ responses were also generated from GPT-3.5, GPT-4, and Bard. LLMs were then prompted to improve the readability of their initial responses. The readability of institutional responses, initial LLM responses, and simplified LLM responses were graded using validated readability formulas. Accuracy and comprehensiveness of initial and simplified LLM responses were also compared. RESULTS: Responses to 66 FAQs were included. All institutional and initial LLM responses had poor readability, with average reading levels ranging from 9th grade to college graduate. Simplified responses from LLMs had significantly improved readability, with reading levels ranging from 6th grade to college freshman. When comparing simplified LLM responses, GPT-4 responses demonstrated the highest readability, with reading levels ranging from 6th to 9th grade. Accuracy was similar between initial and simplified responses from all LLMs. Comprehensiveness was similar between initial and simplified responses from GPT-3.5 and GPT-4. However, 34.8% of Bard's simplified responses were graded as less comprehensive compared to initial. CONCLUSION: Our study highlights the efficacy of LLMs in enhancing the readability of bariatric surgery PEMs. GPT-4 outperformed other models, generating simplified PEMs from 6th to 9th grade reading levels. Unlike GPT-3.5 and GPT-4, Bard's simplified responses were graded as less comprehensive. We advocate for future studies examining the potential role of LLMs as dynamic and personalized sources of PEMs for diverse patient populations of all literacy levels.

Use of Large Language Models to Assess the Likelihood of Epidemics From the Content of Tweets: Infodemiology Study.

BACKGROUND: Previous work suggests that Google searches could be useful in identifying conjunctivitis epidemics. Content-based assessment of social media content may provide additional value in serving as early indicators of conjunctivitis and other systemic infectious diseases. OBJECTIVE: We investigated whether large language models, specifically GPT-3.5 and GPT-4 (OpenAI), can provide probabilistic assessments of whether social media posts about conjunctivitis could indicate a regional outbreak. METHODS: A total of 12,194 conjunctivitis-related tweets were obtained using a targeted Boolean search in multiple languages from India, Guam (United States), Martinique (France), the Philippines, American Samoa (United States), Fiji, Costa Rica, Haiti, and the Bahamas, covering the time frame from January 1, 2012, to March 13, 2023. By providing these tweets via prompts to GPT-3.5 and GPT-4, we obtained probabilistic assessments that were validated by 2 human raters. We then calculated Pearson correlations of these time series with tweet volume and the occurrence of known outbreaks in these 9 locations, with time series bootstrap used to compute CIs. RESULTS: Probabilistic assessments derived from GPT-3.5 showed correlations of 0.60 (95% CI 0.47-0.70) and 0.53 (95% CI 0.40-0.65) with the 2 human raters, with higher results for GPT-4. The weekly averages of GPT-3.5 probabilities showed substantial correlations with weekly tweet volume for 44% (4/9) of the countries, with correlations ranging from 0.10 (95% CI 0.0-0.29) to 0.53 (95% CI 0.39-0.89), with larger correlations for GPT-4. More modest correlations were found for correlation with known epidemics, with substantial correlation only in American Samoa (0.40, 95% CI 0.16-0.81). CONCLUSIONS: These findings suggest that GPT prompting can efficiently assess the content of social media posts and indicate possible disease outbreaks to a degree of accuracy comparable to that of humans. Furthermore, we found that automated content analysis of tweets is related to tweet volume for conjunctivitis-related posts in some locations and to the occurrence of actual epidemics. Future work may improve the sensitivity and specificity of these methods for disease outbreak detection.

Enhanced Artificial Intelligence Strategies in Renal Oncology: Iterative Optimization and Comparative Analysis of GPT 3.5 Versus 4.0.

BACKGROUND: The rise of artificial intelligence (AI) in medicine has revealed the potential of ChatGPT as a pivotal tool in medical diagnosis and treatment. This study assesses the efficacy of ChatGPT versions 3.5 and 4.0 in addressing renal cell carcinoma (RCC) clinical inquiries. Notably, fine-tuning and iterative optimization of the model corrected ChatGPT's limitations in this area. METHODS: In our study, 80 RCC-related clinical questions from urology experts were posed three times to both ChatGPT 3.5 and ChatGPT 4.0, seeking binary (yes/no) responses. We then statistically analyzed the answers. Finally, we fine-tuned the GPT-3.5 Turbo model using these questions, and assessed its training outcomes. RESULTS: We found that the average accuracy rates of answers provided by ChatGPT versions 3.5 and 4.0 were 67.08% and 77.50%, respectively. ChatGPT 4.0 outperformed ChatGPT 3.5, with a higher accuracy rate in responses (p < 0.05). By counting the number of correct responses to the 80 questions, we then found that although ChatGPT 4.0 performed better (p < 0.05), both versions were subject to instability in answering. Finally, by fine-tuning the GPT-3.5 Turbo model, we found that the correct rate of responses to these questions could be stabilized at 93.75%. Iterative optimization of the model can result in 100% response accuracy. CONCLUSION: We compared ChatGPT versions 3.5 and 4.0 in addressing clinical RCC questions, identifying their limitations. By applying the GPT-3.5 Turbo fine-tuned model iterative training method, we enhanced AI strategies in renal oncology. This approach is set to enhance ChatGPT's database and clinical guidance capabilities, optimizing AI in this field.

Can large language models reason about medical questions?

Although large language models often produce impressive outputs, it remains unclear how they perform in real-world scenarios requiring strong reasoning skills and expert domain knowledge. We set out to investigate whether closed- and open-source models (GPT-3.5, Llama 2, etc.) can be applied to answer and reason about difficult real-world-based questions. We focus on three popular medical benchmarks (MedQA-US Medical Licensing Examination [USMLE], MedMCQA, and PubMedQA) and multiple prompting scenarios: chain of thought (CoT; think step by step), few shot, and retrieval augmentation. Based on an expert annotation of the generated CoTs, we found that InstructGPT can often read, reason, and recall expert knowledge. Last, by leveraging advances in prompt engineering (few-shot and ensemble methods), we demonstrated that GPT-3.5 not only yields calibrated predictive distributions but also reaches the passing score on three datasets: MedQA-USMLE (60.2%), MedMCQA (62.7%), and PubMedQA (78.2%). Open-source models are closing the gap: Llama 2 70B also passed the MedQA-USMLE with 62.5% accuracy.

Improving large language models for clinical named entity recognition via prompt engineering.

IMPORTANCE: The study highlights the potential of large language models, specifically GPT-3.5 and GPT-4, in processing complex clinical data and extracting meaningful information with minimal training data. By developing and refining prompt-based strategies, we can significantly enhance the models' performance, making them viable tools for clinical NER tasks and possibly reducing the reliance on extensive annotated datasets. OBJECTIVES: This study quantifies the capabilities of GPT-3.5 and GPT-4 for clinical named entity recognition (NER) tasks and proposes task-specific prompts to improve their performance. MATERIALS AND METHODS: We evaluated these models on 2 clinical NER tasks: (1) to extract medical problems, treatments, and tests from clinical notes in the MTSamples corpus, following the 2010 i2b2 concept extraction shared task, and (2) to identify nervous system disorder-related adverse events from safety reports in the vaccine adverse event reporting system (VAERS). To improve the GPT models' performance, we developed a clinical task-specific prompt framework that includes (1) baseline prompts with task description and format specification, (2) annotation guideline-based prompts, (3) error analysis-based instructions, and (4) annotated samples for few-shot learning. We assessed each prompt's effectiveness and compared the models to BioClinicalBERT. RESULTS: Using baseline prompts, GPT-3.5 and GPT-4 achieved relaxed F1 scores of 0.634, 0.804 for MTSamples and 0.301, 0.593 for VAERS. Additional prompt components consistently improved model performance. When all 4 components were used, GPT-3.5 and GPT-4 achieved relaxed F1 socres of 0.794, 0.861 for MTSamples and 0.676, 0.736 for VAERS, demonstrating the effectiveness of our prompt framework. Although these results trail BioClinicalBERT (F1 of 0.901 for the MTSamples dataset and 0.802 for the VAERS), it is very promising considering few training samples are needed. DISCUSSION: The study's findings suggest a promising direction in leveraging LLMs for clinical NER tasks. However, while the performance of GPT models improved with task-specific prompts, there's a need for further development and refinement. LLMs like GPT-4 show potential in achieving close performance to state-of-the-art models like BioClinicalBERT, but they still require careful prompt engineering and understanding of task-specific knowledge. The study also underscores the importance of evaluation schemas that accurately reflect the capabilities and performance of LLMs in clinical settings. CONCLUSION: While direct application of GPT models to clinical NER tasks falls short of optimal performance, our task-specific prompt framework, incorporating medical knowledge and training samples, significantly enhances GPT models' feasibility for potential clinical applications.

Diminished diversity-of-thought in a standard large language model.

We test whether large language models (LLMs) can be used to simulate human participants in social-science studies. To do this, we ran replications of 14 studies from the Many Labs 2 replication project with OpenAI's text-davinci-003 model, colloquially known as GPT-3.5. Based on our pre-registered analyses, we find that among the eight studies we could analyse, our GPT sample replicated 37.5% of the original results and 37.5% of the Many Labs 2 results. However, we were unable to analyse the remaining six studies due to an unexpected phenomenon we call the "correct answer" effect. Different runs of GPT-3.5 answered nuanced questions probing political orientation, economic preference, judgement, and moral philosophy with zero or near-zero variation in responses: with the supposedly "correct answer." In one exploratory follow-up study, we found that a "correct answer" was robust to changing the demographic details that precede the prompt. In another, we found that most but not all "correct answers" were robust to changing the order of answer choices. One of our most striking findings occurred in our replication of the Moral Foundations Theory survey results, where we found GPT-3.5 identifying as a political conservative in 99.6% of the cases, and as a liberal in 99.3% of the cases in the reverse-order condition. However, both self-reported 'GPT conservatives' and 'GPT liberals' showed right-leaning moral foundations. Our results cast doubts on the validity of using LLMs as a general replacement for human participants in the social sciences. Our results also raise concerns that a hypothetical AI-led future may be subject to a diminished diversity of thought.