Large Language Models (LLMs) have shown significant limitations in understanding creative content, as demonstrated by Hessel et al. (2023)'s influential work on the New Yorker Cartoon Caption Contest (NYCCC). Their study exposed a substantial gap between LLMs and humans in humor comprehension, establishing that understanding and evaluating creative content is key challenge in AI development. We revisit this challenge by decomposing humor understanding into three components and systematically improve each: enhancing visual understanding through improved annotation, utilizing LLM-generated humor reasoning and explanations, and implementing targeted alignment with human preference data. Our refined approach achieves 82.4% accuracy in caption ranking, singificantly improving upon the previous 67% benchmark and matching the performance of world-renowned human experts in this domain. Notably, while attempts to mimic subgroup preferences through various persona prompts showed minimal impact, model finetuning with crowd preferences proved remarkably effective. These findings reveal that LLM limitations in creative judgment can be effectively addressed through focused alignment to specific subgroups and individuals. Lastly, we propose the position that achieving artificial general intelligence necessitates systematic collection of human preference data across creative domains. We advocate that just as human creativity is deeply influenced by individual and cultural preferences, training LLMs with diverse human preference data may be essential for developing true creative understanding.

Guesstimation, the task of making approximate quantity estimates, is a common real-world challenge. However, it has been largely overlooked in large language models (LLMs) and vision language models (VLMs) research. We introduce a novel guesstimation dataset, MARBLES. This dataset requires one to estimate how many items (e.g., marbles) can fit into containers (e.g., a one-cup measuring cup), both with and without accompanying images. Inspired by the social science concept of the ``Wisdom of Crowds'' (WOC) - taking the median from estimates from a crowd), which has proven effective in guesstimation, we propose ``WOC decoding'' strategy for LLM guesstimation. We show that LLMs/VLMs perform well on guesstimation, suggesting that they possess some level of a "world model" necessary for guesstimation. Moreover, similar to human performance, the WOC decoding method improves LLM/VLM guesstimation accuracy. Furthermore, the inclusion of images in the multimodal condition enhances model performance. These results highlight the value of WOC decoding strategy for LLMs/VLMs and position guesstimation as a probe for evaluating LLMs/VLMs' world model. As LLMs' world model is a fundamental prerequisite for many real-world tasks, e.g., human-AI teaming, our findings have broad implications for the AI community.

We present a novel multimodal preference dataset for creative tasks, consisting of over 250 million human ratings on more than 2.2 million captions, collected through crowdsourcing rating data for The New Yorker's weekly cartoon caption contest over the past eight years. This unique dataset supports the development and evaluation of multimodal large language models and preference-based fine-tuning algorithms for humorous caption generation. We propose novel benchmarks for judging the quality of model-generated captions, utilizing both GPT4 and human judgments to establish ranking-based evaluation strategies. Our experimental results highlight the limitations of current fine-tuning methods, such as RLHF and DPO, when applied to creative tasks. Furthermore, we demonstrate that even state-of-the-art models like GPT4 and Claude currently underperform top human contestants in generating humorous captions. As we conclude this extensive data collection effort, we release the entire preference dataset to the research community, fostering further advancements in AI humor generation and evaluation.

Accurately simulating human opinion dynamics is crucial for understanding a variety of societal phenomena, including polarization and the spread of misinformation. However, the agent-based models (ABMs) commonly used for such simulations lack fidelity to human behavior. We propose a new approach to simulating opinion dynamics based on populations of Large Language Models (LLMs). Our findings reveal a strong inherent bias in LLM agents towards accurate information, leading to consensus in line with scientific reality. However, this bias limits the simulation of individuals with resistant views on issues like climate change. After inducing confirmation bias through prompt engineering, we observed opinion fragmentation in line with existing agent-based research. These insights highlight the promise and limitations of LLM agents in this domain and suggest a path forward: refining LLMs with real-world discourse to better simulate the evolution of human beliefs.

This study investigates the potential of Large Language Models (LLMs) to simulate human group dynamics, particularly within politically charged contexts. We replicate the Wisdom of Partisan Crowds phenomenon using LLMs to role-play as Democrat and Republican personas, engaging in a structured interaction akin to human group study. Our approach evaluates how agents' responses evolve through social influence. Our key findings indicate that LLM agents role-playing detailed personas and without Chain-of-Thought (CoT) reasoning closely align with human behaviors, while having CoT reasoning hurts the alignment. However, incorporating explicit biases into agent prompts does not necessarily enhance the wisdom of partisan crowds. Moreover, fine-tuning LLMs with human data shows promise in achieving human-like behavior but poses a risk of overfitting certain behaviors. These findings show the potential and limitations of using LLM agents in modeling human group phenomena.

This paper considers how interactions with AI algorithms can boost human creative thought. We employ a psychological task that demonstrates limits on human creativity, namely semantic feature generation: given a concept name, respondents must list as many of its features as possible. Human participants typically produce only a fraction of the features they know before getting "stuck." In experiments with humans and with a language AI (GPT-4) we contrast behavior in the standard task versus a variant in which participants can ask for algorithmically-generated hints. Algorithm choice is administered by a multi-armed bandit whose reward indicates whether the hint helped generating more features. Humans and the AI show similar benefits from hints, and remarkably, bandits learning from AI responses prefer the same prompting strategy as those learning from human behavior. The results suggest that strategies for boosting human creativity via computer interactions can be learned by bandits run on groups of simulated participants.

Neural network models of language have long been used as a tool for developing hypotheses about conceptual representation in the mind and brain. For many years, such use involved extracting vector-space representations of words and using distances among these to predict or understand human behavior in various semantic tasks. Contemporary large language models (LLMs), however, make it possible to interrogate the latent structure of conceptual representations using experimental methods nearly identical to those commonly used with human participants. The current work utilizes three common techniques borrowed from cognitive psychology to estimate and compare the structure of concepts in humans and a suite of LLMs. In humans, we show that conceptual structure is robust to differences in culture, language, and method of estimation. Structures estimated from LLM behavior, while individually fairly consistent with those estimated from human behavior, vary much more depending upon the particular task used to generate responses--across tasks, estimates of conceptual structure from the very same model cohere less with one another than do human structure estimates. These results highlight an important difference between contemporary LLMs and human cognition, with implications for understanding some fundamental limitations of contemporary machine language.

A central goal in cognitive science is to characterize human knowledge of concepts and their properties. Many have used human-generated feature lists as norms for establishing the structural relationship between concepts in the human mind (McRae et al., 2005; Devereux et al., 2014; De Deyne et al., 2008; Buchanan et al., 2019), but this requires extensive human labor. Large language models (LLMs) have recently shown impressive capabilities when generating properties of objects (Hansen & Hebart, 2022) or answering questions(Ouyang et al., 2022; Brown et al., 2020; Hoffmann et al., 2022; Chowdhery et al., 2022; Wei et al., 2021) and thus suggest an avenue for more efficient characterization of human knowledge structures, but even state-of-the-art models can routinely fail on many common-sense questions of fact. GTP3-davinci, for instance, will deny that alligators are green, while asserting that they can be used to suck dust up from surfaces. Thus, human effort can generate high-quality norms, but with prohibitive costs, while LLMs can produce norms with little human effort, but with considerably less accuracy. This paper considers whether human and machine effort can combine to efficiently estimate high-quality semantic feature vectors.

In cognitive science, efforts to understand the structure of human concepts have relied on semantic feature norms: participants list all the properties they believe to be true of a given concept; responses are collected from many participants for many concepts; overlap in the resulting feature vectors captures the degree to which concepts are semantically related(Rosch, 1973; McRae et al., 2005). Yet participants often produce only a fraction of what they know for each concept: tigers have DNA, can breathe, and are alive, but these properties are not typically produced in feature norms for tiger. Such omissions are important because they express deep conceptual structure: having DNA and breathing connect tigers to all other plants and animals. To better capture such structure, some studies ask human participants to make yes/no judgments for all possible properties across every concept. Thus if "can breathe" was listed for a single concept, human raters would then evaluate whether each other concept in the dataset can breathe. This verification step significantly enriches the conceptual structure that features norms express (De Deyne et al., 2008), but is exceedingly costly in human labor: the number of verification questions asked increases exponentially with the number of concepts probed. Previous work has shown that the conceptual structure of a large language model (LLM) for semantic feature listing is similar to human conceptual structure (Suresh et al., 2023; Bhatia & Richie, 2022). In this paper we consider whether this step can be reliably "outsourced" to an open sourced LLM optimized for question-answering, specifically the opensource FLAN-T5 XXL model (Chung et al., 2022; Wei et al., 2021), focusing on two questions: (1) How accurately does the LLM capture human responses to the questions?