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.

This article describes how technical infrastructure developed by the nonprofit OpenMined enables external scrutiny of AI systems without compromising sensitive information. Independent external scrutiny of AI systems provides crucial transparency into AI development, so it should be an integral component of any approach to AI governance. In practice, external researchers have struggled to gain access to AI systems because of AI companies' legitimate concerns about security, privacy, and intellectual property. But now, privacy-enhancing technologies (PETs) have reached a new level of maturity: end-to-end technical infrastructure developed by OpenMined combines several PETs into various setups that enable privacy-preserving audits of AI systems. We showcase two case studies where this infrastructure has been deployed in real-world governance scenarios: "Understanding Social Media Recommendation Algorithms with the Christchurch Call" and "Evaluating Frontier Models with the UK AI Safety Institute." We describe types of scrutiny of AI systems that could be facilitated by current setups and OpenMined's proposed future setups. We conclude that these innovative approaches deserve further exploration and support from the AI governance community. Interested policymakers can focus on empowering researchers on a legal level.

Distinguishing in- and out-of-distribution (OOD) inputs is crucial for reliable deployment of classification systems. However, OOD data is typically unavailable or difficult to collect, posing a significant challenge for accurate OOD detection. In this work, we present a method that harnesses the generative capabilities of Large Language Models (LLMs) to create high-quality synthetic OOD proxies, eliminating the dependency on any external OOD data source. We study the efficacy of our method on classical text classification tasks such as toxicity detection and sentiment classification as well as classification tasks arising in LLM development and deployment, such as training a reward model for RLHF and detecting misaligned generations. Extensive experiments on nine InD-OOD dataset pairs and various model sizes show that our approach dramatically lowers false positive rates (achieving a perfect zero in some cases) while maintaining high accuracy on in-distribution tasks, outperforming baseline methods by a significant margin.

When deploying large language models (LLMs), it is important to ensure that these models are not only capable, but also reliable. Many benchmarks have been created to track LLMs' growing capabilities, however there has been no similar focus on measuring their reliability. To understand the potential ramifications of this gap, we investigate how well current benchmarks quantify model reliability. We find that pervasive label errors can compromise these evaluations, obscuring lingering model failures and hiding unreliable behavior. Motivated by this gap in the evaluation of reliability, we then propose the concept of so-called platinum benchmarks, i.e., benchmarks carefully curated to minimize label errors and ambiguity. As a first attempt at constructing such benchmarks, we revise examples from fifteen existing popular benchmarks. We evaluate a wide range of models on these platinum benchmarks and find that, indeed, frontier LLMs still exhibit failures on simple tasks such as elementary-level math word problems. Analyzing these failures further reveals previously unidentified patterns of problems on which frontier models consistently struggle. We provide code at https://github.com/MadryLab/platinum-benchmarks

Text-embedding models often exhibit biases arising from the data on which they are trained. In this paper, we examine a hitherto unexplored bias in text-embeddings: bias arising from the presence of $\textit{names}$ such as persons, locations, organizations etc. in the text. Our study shows how the presence of $\textit{name-bias}$ in text-embedding models can potentially lead to erroneous conclusions in assessment of thematic similarity.Text-embeddings can mistakenly indicate similarity between texts based on names in the text, even when their actual semantic content has no similarity or indicate dissimilarity simply because of the names in the text even when the texts match semantically. We first demonstrate the presence of name bias in different text-embedding models and then propose $\textit{text-anonymization}$ during inference which involves removing references to names, while preserving the core theme of the text. The efficacy of the anonymization approach is demonstrated on two downstream NLP tasks, achieving significant performance gains. Our simple and training-optimization-free approach offers a practical and easily implementable solution to mitigate name bias.

Large Language Models (LLMs) are increasingly being deployed on edge devices for long-context settings, creating a growing need for fast and efficient long-context inference. In these scenarios, the Key-Value (KV) cache is the primary bottleneck in terms of both GPU memory and latency, as the full KV cache must be loaded for each decoding step. While speculative decoding is a widely accepted technique to accelerate autoregressive decoding, existing methods often struggle to achieve significant speedups due to inefficient KV cache optimization strategies and result in low acceptance rates. To address these challenges, we propose a novel self-speculative decoding framework, QuantSpec, where the draft model shares the architecture of the target model but employs a hierarchical 4-bit quantized KV cache and 4-bit quantized weights for acceleration. QuantSpec maintains high acceptance rates ($>$90%) and reliably provides consistent end-to-end speedups upto $\sim2.5\times$, outperforming other self-speculative decoding methods that use sparse KV cache for long-context LLM inference. QuantSpec also reduces the memory requirements by $\sim 1.3\times$ compared to these alternatives.

Despite the much proliferation of AI ethical principles in recent years, there is a challenge of assuring AI ethics with current AI ethics frameworks in real-world applications. While system safety has emerged as a distinct discipline for a long time, originated from safety concerns in early aircraft manufacturing. The safety assurance is now an indispensable component in safety critical domains. Motivated by the assurance approaches for safety-critical systems such as aviation, this paper introduces the concept of AI ethics assurance cases into the AI ethics assurance. Three pillars of user requirements, evidence, and validation are proposed as key components and integrated into AI ethics assurance cases for a new approach of user requirements-oriented AI ethics assurance. The user requirements-oriented AI ethics assurance case is set up based on three pillars and hazard analysis methods used in the safety assurance of safety-critical systems. This paper also proposes a platform named Ethical-Lens (E-LENS) to implement the user requirements-oriented AI ethics assurance approach. The proposed user requirements-based E-LENS platform is then applied to assure AI ethics of an AI-driven human resource shortlisting system as a case study to show the effectiveness of the proposed approach.

With the rise of large multi-modal AI models, fuelled by recent interest in large language models (LLMs), the notion of artificial general intelligence (AGI) went from being restricted to a fringe community, to dominate mainstream large AI development programs. In contrast, in this paper, we make a \emph{case for specialisation}, by reviewing the pitfalls of generality and stressing the industrial value of specialised systems. Our contribution is threefold. First, we review the most widely accepted arguments \emph{against} specialisation, and discuss how their relevance in the context of human labour is actually an argument \emph{for} specialisation in the case of non human agents, be they algorithms or human organisations. Second, we propose four arguments \emph{in favor of} specialisation, ranging from machine learning robustness, to computer security, social sciences and cultural evolution. Third, we finally make a case for \emph{specification}, discuss how the machine learning approach to AI has so far failed to catch up with good practices from safety-engineering and formal verification of software, and discuss how some emerging good practices in machine learning help reduce this gap. In particular, we justify the need for \emph{specified governance} for hard-to-specify systems.

Google has made one of the most substantive changes to its AI principles since first publishing them in 2018. In a change spotted by The Washington Post, the search giant edited the document to remove pledges it had made promising it would not "design or deploy" AI tools for use in weapons or surveillance technology. Previously, those guidelines included a section titled "applications we will not pursue," which is not present in the current version of the document. Instead, there's now a section titled "responsible development and deployment." There, Google says it will implement "appropriate human oversight, due diligence, and feedback mechanisms to align with user goals, social responsibility, and widely accepted principles of international law and human rights." That's a far broader commitment than the specific ones the company made as recently as the end of last month when the prior version of its AI principles was still live on its website.

Shares of Google's parent company Alphabet fell more than 6% after the company reported a slight miss in expected revenue on Tuesday. The company reported 96.5bn, compared with analyst expectations of 96.67 bn. The company surpassed investors' expectations of 2.13 in earnings per share, however, with 2.15 in EPS. "Q4 was a strong quarter driven by our leadership in AI and momentum across the business," Alphabet chief executive Sundar Pichai wrote in a statement. "We are building, testing, and launching products and models faster than ever, and making significant progress in compute and driving efficiencies."