Robert West

Martin Josifoski, Lars Klein, Maxime Peyrard et al.

Recent advances in artificial intelligence (AI) have produced highly capable and controllable systems. This creates unprecedented opportunities for structured reasoning as well as collaboration among multiple AI systems and humans. To fully realize this potential, it is essential to develop a principled way of designing and studying such structured interactions. For this purpose, we introduce the conceptual framework Flows. Flows are self-contained building blocks of computation, with an isolated state, communicating through a standardized message-based interface. This modular design simplifies the process of creating Flows by allowing them to be recursively composed into arbitrarily nested interactions and is inherently concurrency-friendly. Crucially, any interaction can be implemented using this framework, including prior work on AI-AI and human-AI interactions, prompt engineering schemes, and tool augmentation. We demonstrate the potential of Flows on competitive coding, a challenging task on which even GPT-4 struggles. Our results suggest that structured reasoning and collaboration substantially improve generalization, with AI-only Flows adding +21 and human-AI Flows adding +54 absolute points in terms of solve rate. To support rapid and rigorous research, we introduce the aiFlows library embodying Flows. The aiFlows library is available at https://github.com/epfl-dlab/aiflows. Data and Flows for reproducing our experiments are available at https://github.com/epfl-dlab/cc_flows.

Luca Mouchel, Debjit Paul, Shaobo Cui et al.

Despite the remarkable performance of Large Language Models (LLMs) in natural language processing tasks, they still struggle with generating logically sound arguments, resulting in potential risks such as spreading misinformation. To address this issue, we introduce FIPO, a fallacy-informed framework that leverages preference optimization methods to steer LLMs toward logically sound arguments. FIPO includes a classification loss, to capture the fine-grained information on fallacy types. Our results on argumentation datasets show that our method reduces the fallacy errors by up to 17.5%. Furthermore, our human evaluation results indicate that the quality of the generated arguments by our method significantly outperforms the fine-tuned baselines, as well as other preference optimization methods, such as DPO. These findings highlight the importance of ensuring models are aware of logical fallacies for effective argument generation. Our code is available at github.com/lucamouchel/Logical-Fallacies.

Debjit Paul, Robert West, Antoine Bosselut et al.

Large language models (LLMs) have been shown to perform better when asked to reason step-by-step before answering a question. However, it is unclear to what degree the model's final answer is faithful to the stated reasoning steps. In this paper, we perform a causal mediation analysis on twelve LLMs to examine how intermediate reasoning steps generated by the LLM influence the final outcome and find that LLMs do not reliably use their intermediate reasoning steps when generating an answer. To address this issue, we introduce FRODO, a framework to tailor small-sized LMs to generate correct reasoning steps and robustly reason over these steps. FRODO consists of an inference module that learns to generate correct reasoning steps using an implicit causal reward function and a reasoning module that learns to faithfully reason over these intermediate inferences using a counterfactual and causal preference objective. Our experiments show that FRODO significantly outperforms four competitive baselines. Furthermore, FRODO improves the robustness and generalization ability of the reasoning LM, yielding higher performance on out-of-distribution test sets. Finally, we find that FRODO's rationales are more faithful to its final answer predictions than standard supervised fine-tuning.

Francesco Salvi, Giuseppe Russo, Adam Barla et al.

Meat consumption is a major driver of global greenhouse gas emissions. While pricing interventions have shown potential to reduce meat intake, previous studies have focused on highly constrained environments with limited consumer choice. Here, we present the first large-scale field experiment to evaluate multiple pricing interventions in a real-world, competitive setting. Using a sequential crossover design with matched menus in a Swiss university campus, we systematically compared vegetarian-meal discounts (-2.5 CHF), meat surcharges (+2.5 CHF), and a combined scheme (-1.2 CHF=+1.2 CHF) across four campus cafeterias. Only the surcharge and combined interventions led to significant increases in vegetarian meal uptake--by 26.4% and 16.6%, respectively--and reduced CO2 emissions per meal by 7.4% and 11.3%, respectively. The surcharge, while effective, triggered a 12.3% drop in sales at intervention sites and a corresponding 14.9% increase in non-treated locations, hence causing a spillover effect that completely offset environmental gains. In contrast, the combined approach achieved meaningful emission reductions without significant effects on overall sales or revenue, making it both effective and economically viable. Notably, pricing interventions were equally effective for both vegetarian-leaning customers and habitual meat-eaters, stimulating change even within entrenched dietary habits. Our results show that balanced pricing strategies can reduce the carbon footprint of realistic food environments, but require coordinated implementation to maximize climate benefits and avoid unintended spillover effects.

Tim R. Davidson, Veniamin Veselovsky, Martin Josifoski et al. · stanford

We introduce an approach to evaluate language model (LM) agency using negotiation games. This approach better reflects real-world use cases and addresses some of the shortcomings of alternative LM benchmarks. Negotiation games enable us to study multi-turn, and cross-model interactions, modulate complexity, and side-step accidental evaluation data leakage. We use our approach to test six widely used and publicly accessible LMs, evaluating performance and alignment in both self-play and cross-play settings. Noteworthy findings include: (i) only closed-source models tested here were able to complete these tasks; (ii) cooperative bargaining games proved to be most challenging to the models; and (iii) even the most powerful models sometimes "lose" to weaker opponents

Viacheslav Surkov, Chris Wendler, Antonio Mari et al.

For large language models (LLMs), sparse autoencoders (SAEs) have been shown to decompose intermediate representations that often are not interpretable directly into sparse sums of interpretable features, facilitating better control and subsequent analysis. However, similar analyses and approaches have been lacking for text-to-image models. We investigate the possibility of using SAEs to learn interpretable features for SDXL Turbo, a few-step text-to-image diffusion model. To this end, we train SAEs on the updates performed by transformer blocks within SDXL Turbo's denoising U-net in its 1-step setting. Interestingly, we find that they generalize to 4-step SDXL Turbo and even to the multi-step SDXL base model (i.e., a different model) without additional training. In addition, we show that their learned features are interpretable, causally influence the generation process, and reveal specialization among the blocks. We do so by creating RIEBench, a representation-based image editing benchmark, for editing images while they are generated by turning on and off individual SAE features. This allows us to track which transformer blocks' features are the most impactful depending on the edit category. Our work is the first investigation of SAEs for interpretability in text-to-image diffusion models and our results establish SAEs as a promising approach for understanding and manipulating the internal mechanisms of text-to-image models.

Giuseppe Russo, Kristina Gligorić, Vincent Moreau et al.

Reducing meat consumption is crucial for achieving global environmental and nutritional targets. Meat-Free Day (MFD) is a widely adopted strategy to address this challenge by encouraging plant-based diets through the removal of animal-based meals. We assessed the environmental, behavioral, and nutritional impacts of MFD by implementing 67 MFDs over 18 months (once a week on a randomly chosen day) across 12 cafeterias on a large university campus, analyzing over 400,000 food purchases. MFD reduced on-campus food-related greenhouse gas (GHG) emissions on treated days by 52.9% and contributed to improved fiber (+26.9%) and cholesterol (-4.5%) consumption without altering caloric intake. These nutritional benefits were, however, accompanied by a 27.6% decrease in protein intake and a 34.2% increase in sugar consumption. Moreover, the increase in plant-based meals did not carry over to subsequent days, as evidenced by a 3.5% rebound in animal-based meal consumption on days immediately following treated days. MFD also led to a 16.8% drop in on-campus meal sales on treated days.Monte Carlo simulations suggest that if 8.7% of diners were to eat burgers off-campus on treated days, MFD's GHG savings would be fully negated. As our analysis identifies on-campus customer retention as the main challenge to MFD effectiveness, we recommend combining MFD with customer retention interventions to ensure environmental and nutritional benefits.

Andreea Nica, Ivan Zakazov, Nicolas Mario Baldwin et al.

Prompt optimization improves the reasoning abilities of large language models (LLMs) without requiring parameter updates to the target model. Following heuristic-based "Think step by step" approaches, the field has evolved in two main directions: while one group of methods uses textual feedback to elicit improved prompts from general-purpose LLMs in a training-free way, a concurrent line of research relies on numerical rewards to train a special prompt model, tailored for providing optimal prompts to the target model. In this paper, we introduce the Textual Reward Prompt framework (TRPrompt), which unifies these approaches by directly incorporating textual feedback into training of the prompt model. Our framework does not require prior dataset collection and is being iteratively improved with the feedback on the generated prompts. When coupled with the capacity of an LLM to internalize the notion of what a "good" prompt is, the high-resolution signal provided by the textual rewards allows us to train a prompt model yielding state-of-the-art query-specific prompts for the problems from the challenging math datasets GSMHard and MATH.

Jason Becker, Chris Wendler, Peter Baylies et al.

Instead of performing text-conditioned denoising in the image domain, latent diffusion models (LDMs) operate in latent space of a variational autoencoder (VAE), enabling more efficient processing at reduced computational costs. However, while the diffusion process has moved to the latent space, the contrastive language-image pre-training (CLIP) models, as used in many image processing tasks, still operate in pixel space. Doing so requires costly VAE-decoding of latent images before they can be processed. In this paper, we introduce Latent-CLIP, a CLIP model that operates directly in the latent space. We train Latent-CLIP on 2.7B pairs of latent images and descriptive texts, and show that it matches zero-shot classification performance of similarly sized CLIP models on both the ImageNet benchmark and a LDM-generated version of it, demonstrating its effectiveness in assessing both real and generated content. Furthermore, we construct Latent-CLIP rewards for reward-based noise optimization (ReNO) and show that they match the performance of their CLIP counterparts on GenEval and T2I-CompBench while cutting the cost of the total pipeline by 21%. Finally, we use Latent-CLIP to guide generation away from harmful content, achieving strong performance on the inappropriate image prompts (I2P) benchmark and a custom evaluation, without ever requiring the costly step of decoding intermediate images.

Dimitrios Rontogiannis, Maxime Peyrard, Nicolas Baldwin et al.

Standard single-turn, static benchmarks fall short in evaluating the nuanced capabilities of Large Language Models (LLMs) on complex tasks such as software engineering. In this work, we propose a novel interactive evaluation framework that assesses LLMs on multi-requirement programming tasks through structured, feedback-driven dialogue. Each task is modeled as a requirement dependency graph, and an ``interviewer'' LLM, aware of the ground-truth solution, provides minimal, targeted hints to an ``interviewee'' model to help correct errors and fulfill target constraints. This dynamic protocol enables fine-grained diagnostic insights into model behavior, uncovering strengths and systematic weaknesses that static benchmarks fail to measure. We build on DevAI, a benchmark of 55 curated programming tasks, by adding ground-truth solutions and evaluating the relevance and utility of interviewer hints through expert annotation. Our results highlight the importance of dynamic evaluation in advancing the development of collaborative code-generating agents.

Robert L. West, Spencer Eckler, Brendan Conway-Smith et al.

This article presents a theoretical framework for adapting the Common Model of Cognition to large generative network models within the field of artificial intelligence. This can be accomplished by restructuring modules within the Common Model into shadow production systems that are peripheral to a central production system, which handles higher-level reasoning based on the shadow productions' output. Implementing this novel structure within the Common Model allows for a seamless connection between cognitive architectures and generative neural networks.

Brendan Conway-Smith, Robert L. West

There have been increasing challenges to dual-system descriptions of System-1 and System-2, critiquing them as imprecise and fostering misconceptions. We address these issues here by way of Dennett's appeal to use computational thinking as an analytical tool, specifically we employ the Common Model of Cognition. Results show that the characteristics thought to be distinctive of System-1 and System-2 instead form a spectrum of cognitive properties. By grounding System-1 and System-2 in the Common Model we aim to clarify their underlying mechanisms, persisting misconceptions, and implications for metacognition.

Brendan Conway-Smith, Robert L. West

Attempts to import dual-system descriptions of System-1 and System-2 into AI have been hindered by a lack of clarity over their distinction. We address this and other issues by situating System-1 and System-2 within the Common Model of Cognition. Results show that what are thought to be distinctive characteristics of System-1 and 2 instead form a spectrum of cognitive properties. The Common Model provides a comprehensive vision of the computational units involved in System-1 and System-2, their underlying mechanisms, and the implications for learning, metacognition, and emotion.

Sylvain Lugeon, Tiziano Piccardi, Robert West

Currently, publicly available models for website classification do not offer an embedding method and have limited support for languages beyond English. We release a dataset of more than two million category-labeled websites in 92 languages collected from Curlie, the largest multilingual human-edited Web directory. The dataset contains 14 website categories aligned across languages. Alongside it, we introduce Homepage2Vec, a machine-learned pre-trained model for classifying and embedding websites based on their homepage in a language-agnostic way. Homepage2Vec, thanks to its feature set (textual content, metadata tags, and visual attributes) and recent progress in natural language representation, is language-independent by design and generates embedding-based representations. We show that Homepage2Vec correctly classifies websites with a macro-averaged F1-score of 0.90, with stable performance across low- as well as high-resource languages. Feature analysis shows that a small subset of efficiently computable features suffices to achieve high performance even with limited computational resources. We make publicly available the curated Curlie dataset aligned across languages, the pre-trained Homepage2Vec model, and libraries

Chad Peters, Babak Esfandiari, Mohamad Zalat et al.

Learning from Observation (LfO), also known as Behavioral Cloning, is an approach for building software agents by recording the behavior of an expert (human or artificial) and using the recorded data to generate the required behavior. jLOAF is a platform that uses Case-Based Reasoning to achieve LfO. In this paper we interface jLOAF with the popular OpenAI Gym environment. Our experimental results show how our approach can be used to provide a baseline for comparison in this domain, as well as identify the strengths and weaknesses when dealing with environmental complexity.

Ramtin Yazdanian, Leila Zia, Jonathan Morgan et al.

Every day, thousands of users sign up as new Wikipedia contributors. Once joined, these users have to decide which articles to contribute to, which users to seek out and learn from or collaborate with, etc. Any such task is a hard and potentially frustrating one given the sheer size of Wikipedia. Supporting newcomers in their first steps by recommending articles they would enjoy editing or editors they would enjoy collaborating with is thus a promising route toward converting them into long-term contributors. Standard recommender systems, however, rely on users' histories of previous interactions with the platform. As such, these systems cannot make high-quality recommendations to newcomers without any previous interactions -- the so-called cold-start problem. The present paper addresses the cold-start problem on Wikipedia by developing a method for automatically building short questionnaires that, when completed by a newly registered Wikipedia user, can be used for a variety of purposes, including article recommendations that can help new editors get started. Our questionnaires are constructed based on the text of Wikipedia articles as well as the history of contributions by the already onboarded Wikipedia editors. We assess the quality of our questionnaire-based recommendations in an offline evaluation using historical data, as well as an online evaluation with hundreds of real Wikipedia newcomers, concluding that our method provides cohesive, human-readable questions that perform well against several baselines. By addressing the cold-start problem, this work can help with the sustainable growth and maintenance of Wikipedia's diverse editor community.

Robert West, Eric Horvitz

Humor is an essential human trait. Efforts to understand humor have called out links between humor and the foundations of cognition, as well as the importance of humor in social engagement. As such, it is a promising and important subject of study, with relevance for artificial intelligence and human-computer interaction. Previous computational work on humor has mostly operated at a coarse level of granularity, e.g., predicting whether an entire sentence, paragraph, document, etc., is humorous. As a step toward deep understanding of humor, we seek fine-grained models of attributes that make a given text humorous. Starting from the observation that satirical news headlines tend to resemble serious news headlines, we build and analyze a corpus of satirical headlines paired with nearly identical but serious headlines. The corpus is constructed via Unfun.me, an online game that incentivizes players to make minimal edits to satirical headlines with the goal of making other players believe the results are serious headlines. The edit operations used to successfully remove humor pinpoint the words and concepts that play a key role in making the original, satirical headline funny. Our analysis reveals that the humor tends to reside toward the end of headlines, and primarily in noun phrases, and that most satirical headlines follow a certain logical pattern, which we term false analogy. Overall, this paper deepens our understanding of the syntactic and semantic structure of satirical news headlines and provides insights for building humor-producing systems.