Preference alignment via reward models helps build safe, helpful, and reliable large language models (LLMs). However, subjectivity in preference judgments and the lack of representative sampling in preference data collection can introduce new biases, hindering reward models' fairness and equity. In this work, we introduce a framework for evaluating dialect biases in reward models and conduct a case study on biases against African American Language (AAL) through several experiments comparing reward model preferences and behavior on paired White Mainstream English (WME) and both machine-translated and human-written AAL corpora. We show that reward models are less aligned with human preferences when processing AAL texts vs. WME ones (-4\% accuracy on average), frequently disprefer AAL-aligned texts vs. WME-aligned ones, and steer conversations toward WME, even when prompted with AAL texts. Our findings provide a targeted analysis of anti-AAL biases at a relatively understudied stage in LLM development, highlighting representational harms and ethical questions about the desired behavior of LLMs concerning AAL.

Privacy protection of users' entire contribution of samples is important in distributed systems. The most effective approach is the two-stage scheme, which finds a small interval first and then gets a refined estimate by clipping samples into the interval. However, the clipping operation induces bias, which is serious if the sample distribution is heavy-tailed. Besides, users with large local sample sizes can make the sensitivity much larger, thus the method is not suitable for imbalanced users. Motivated by these challenges, we propose a Huber loss minimization approach to mean estimation under user-level differential privacy. The connecting points of Huber loss can be adaptively adjusted to deal with imbalanced users. Moreover, it avoids the clipping operation, thus significantly reducing the bias compared with the two-stage approach. We provide a theoretical analysis of our approach, which gives the noise strength needed for privacy protection, as well as the bound of mean squared error. The result shows that the new method is much less sensitive to the imbalance of user-wise sample sizes and the tail of sample distributions. Finally, we perform numerical experiments to validate our theoretical analysis.

We evaluate how well LLMs understand African American Language (AAL) in comparison to their performance on White Mainstream English (WME), the encouraged "standard" form of English taught in American classrooms. We measure LLM performance using automatic metrics and human judgments for two tasks: a counterpart generation task, where a model generates AAL (or WME) given WME (or AAL), and a masked span prediction (MSP) task, where models predict a phrase that was removed from their input. Our contributions include: (1) evaluation of six pre-trained, large language models on the two language generation tasks; (2) a novel dataset of AAL text from multiple contexts (social media, hip-hop lyrics, focus groups, and linguistic interviews) with human-annotated counterparts in WME; and (3) documentation of model performance gaps that suggest bias and identification of trends in lack of understanding of AAL features.

While the celebrated Word2Vec technique yields semantically rich representations for individual words, there has been relatively less success in extending to generate unsupervised sentences or documents embeddings. Recent work has demonstrated that a distance measure between documents called \emph{Word Mover's Distance} (WMD) that aligns semantically similar words, yields unprecedented KNN classification accuracy. However, WMD is expensive to compute, and it is hard to extend its use beyond a KNN classifier. In this paper, we propose the \emph{Word Mover's Embedding } (WME), a novel approach to building an unsupervised document (sentence) embedding from pre-trained word embeddings. In our experiments on 9 benchmark text classification datasets and 22 textual similarity tasks, the proposed technique consistently matches or outperforms state-of-the-art techniques, with significantly higher accuracy on problems of short length.

Influential has launched raised $12 million to expand its Social Intelligence platform to find influencers for brands with the help of artificial intelligence. The idea is to match influencers with brands who want to team up with the hottest social media attractions. Using IBM's Watson AI tech, Influential wants to make influencer marketing reach scale on the level that brand marketers need to get measurable results. Beverly Hills, California-based Influential uses three different application programming interfaces for Watson to predict whether Fortune 1000 brands will succeed with particular influencer-based marketing campaigns. Influential's Social Intelligence technology examines factors such as psychographic and contextual relevance for brands to identify their audience, profile, and personality based on social media analysis.

Influential announced today that it has raised $12 million in Series B funding. The funding came from existing investors Capital Zed, ECA Ventures, Paradigm Talent Agency, ROAR and Tech Coast Angels, as well as from Hollywood agency WME . Just a couple weeks ago, Influential said it was working with (and had raised money from) WME. The agency is the first to try out a new Influential product called Talent Pro, which gives agents access to social data around a broader pool of talent. Influential founder and CEO Ryan Detert said the product will allow WME -- and, in the future, other agencies -- to sweeten endorsement and promotional deals with more data and to "take an A-list celebrity… and now surround that person with 10 lookalike influencers who are not celebrities themselves." One of Influential's big selling points is its use of artificial intelligence (it's a developer partner with IBM Watson) to help brands and marketers find influencers who would be a good fit for their campaigns.