Search for: All records

Diversity in training data, architecture, and providers is assumed to mitigate homogeneity in LLMs. However, we lack empirical evidence on whether different LLMs differ \textit{meaningfully}. We conduct a large-scale empirical evaluation on over 350 LLMs overall, using two popular leaderboards and a resume-screening task. We find substantial correlation in model errors---on one leaderboard dataset, models agree 60% of the time when both models err. We identify factors driving model correlation, including shared architectures and providers. Crucially, however, larger and more accurate models have highly correlated errors, even with distinct architectures and providers. Finally, we show the effects of correlation in two downstream tasks: LLM-as-judge evaluation and hiring---the latter reflecting theoretical predictions regarding algorithmic monoculture. 

Integrating spatial context into large language models (LLMs) has the potential to revolutionize human-computer interaction, particularly in wearable devices. In this work, we present a novel system architecture that incorporates spatial speech understanding into LLMs, enabling contextually aware and adaptive applications for wearable technologies. Our approach leverages microstructure-based spatial sensing to extract precise Direction of Arrival (DoA) information using a monaural microphone. To address the lack of existing dataset for microstructure-assisted speech recordings, we synthetically create a dataset called OmniTalk by using the LibriSpeech dataset. This spatial information is fused with linguistic embeddings from OpenAI’s Whisper model, allowing each modality to learn complementary contextual representations. The fused embeddings are aligned with the input space of LLaMA-3.2 3B model and fine-tuned with lightweight adaptation technique LoRA to optimize for on-device processing. 

We describe HypotheSAEs, a general method to hypothesize interpretable relationships between text data (e.g., headlines) and a target variable (e.g., clicks). HypotheSAEs has three steps: (1) train a sparse autoencoder on text embeddings to produce interpretable features describing the data distribution, (2) select features that predict the target variable, and (3) generate a natural language interpretation of each feature (e.g., mentions being surprised or shocked) using an LLM. Each interpretation serves as a hypothesis about what predicts the target variable. Compared to baselines, our method better identifies reference hypotheses on synthetic datasets (at least +0.06 in F1) and produces more predictive hypotheses on real datasets (~twice as many significant findings), despite requiring 1-2 orders of magnitude less compute than recent LLM-based methods. HypotheSAEs also produces novel discoveries on two well-studied tasks: explaining partisan differences in Congressional speeches and identifying drivers of engagement with online headlines. 

Online marketplaces use rating systems to promote the discovery of high-quality products. However, these systems also lead to high variance in producers' economic outcomes: a new producer who sells high-quality items, may unluckily receive a low rating early, severely impacting their future popularity. We investigate the design of rating systems that balance the goals of identifying high-quality products (``efficiency'') and minimizing the variance in outcomes of producers of similar quality (individual ``producer fairness'').We show that there is a trade-off between these two goals: rating systems that promote efficiency are necessarily less individually fair to producers. We introduce prior-weighted rating systems as an approach to managing this trade-off. Informally, the system we propose sets a system-wide prior for the quality of an incoming product; subsequently, the system updates that prior to a posterior for each product's quality based on user-generated ratings over time. We show theoretically that in markets where products accrue reviews at an equal rate, the strength of the rating system's prior determines the operating point on the identified trade-off: the stronger the prior, the more the marketplace discounts early ratings data (increasing individual fairness), but the slower the platform is in learning about true item quality (so efficiency suffers). We further analyze this trade-off in a responsive market where customers make decisions based on historical ratings. Through calibrated simulations in 19 different real-world datasets sourced from large online platforms, we show that the choice of prior strength mediates the same efficiency-consistency trade-off in this setting. Overall, we demonstrate that by tuning the prior as a design choice in a prior-weighted rating system, platforms can be intentional about the balance between efficiency and producer fairness. 

The gold standard in human-AI collaboration is complementarity: when combined performance exceeds both the human and algorithm alone. We investigate this challenge in binary classification settings where the goal is to maximize 0-1 accuracy. Given two or more agents who can make calibrated probabilistic predictions, we show a No Free Lunch-style result. Any deterministic collaboration strategy (a function mapping calibrated probabilities into binary classifications) that does not essentially always defer to the same agent will sometimes perform worse than the least accurate agent. In other words, complementarity cannot be achieved for free. The result does suggest one model of collaboration with guarantees, where one agent identifies obvious errors of the other agent. We also use the result to understand the necessary conditions enabling the success of other collaboration techniques, providing guidance to human-AI collaboration.