Language models (LMs) can "memorize" information, i.e., encode training data in their weights in such a way that inference-time queries can lead to verbatim regurgitation of that data. This ability to extract training data can be problematic, for example, when data are private or sensitive. In this work, we investigate methods to mitigate memorization: three regularizer-based, three finetuning-based, and eleven machine unlearning-based methods, with five of the latter being new methods that we introduce. We also introduce TinyMem, a suite of small, computationally-efficient LMs for the rapid development and evaluation of memorization-mitigation methods. We demonstrate that the mitigation methods that we develop using TinyMem can successfully be applied to production-grade LMs, and we determine via experiment that: regularizer-based mitigation methods are slow and ineffective at curbing memorization; fine-tuning-based methods are effective at curbing memorization, but overly expensive, especially for retaining higher accuracies; and unlearning-based methods are faster and more effective, allowing for the precise localization and removal of memorized information from LM weights prior to inference. We show, in particular, that our proposed unlearning method BalancedSubnet outperforms other mitigation methods at removing memorized information while preserving performance on target tasks.

It has been reported that LLMs can recognize their own writing. As this has potential implications for AI safety, yet is relatively understudied, we investigate the phenomenon, seeking to establish whether it robustly occurs at the behavioral level, how the observed behavior is achieved, and whether it can be controlled. First, we find that the Llama3-8b-Instruct chat model - but not the base Llama3-8b model - can reliably distinguish its own outputs from those of humans, and present evidence that the chat model is likely using its experience with its own outputs, acquired during post-training, to succeed at the writing recognition task. Second, we identify a vector in the residual stream of the model that is differentially activated when the model makes a correct self-written-text recognition judgment, show that the vector activates in response to information relevant to self-authorship, present evidence that the vector is related to the concept of "self" in the model, and demonstrate that the vector is causally related to the model's ability to perceive and assert self-authorship. Finally, we show that the vector can be used to control both the model's behavior and its perception, steering the model to claim or disclaim authorship by applying the vector to the model's output as it generates it, and steering the model to believe or disbelieve it wrote arbitrary texts by applying the vector to them as the model reads them.

Alert fatigue is a common issue faced by software teams using the DevSecOps paradigm. The overwhelming number of warnings and alerts generated by security and code scanning tools, particularly in smaller teams where resources are limited, leads to desensitization and diminished responsiveness to security warnings, potentially exposing systems to vulnerabilities. This paper explores the potential of LLMs in generating actionable security reports that emphasize the financial impact and consequences of detected security issues, such as credential leaks, if they remain unaddressed. A survey conducted among developers indicates that LLM-generated reports significantly enhance the likelihood of immediate action on security issues by providing clear, comprehensive, and motivating insights. Integrating these reports into DevSecOps workflows can mitigate attention saturation and alert fatigue, ensuring that critical security warnings are addressed effectively.

In an era of "moving fast and breaking things", regulators have moved slowly to pick up the safety, bias, and legal pieces left in the wake of broken Artificial Intelligence (AI) deployment. Since AI models, such as large language models, are able to push misinformation and stoke division within our society, it is imperative for regulators to employ a framework that mitigates these dangers and ensures user safety. While there is much-warranted discussion about how to address the safety, bias, and legal woes of state-of-the-art AI models, the number of rigorous and realistic mathematical frameworks to regulate AI safety is lacking. We take on this challenge, proposing an auction-based regulatory mechanism that provably incentivizes model-building agents (i) to deploy safer models and (ii) to participate in the regulation process. We provably guarantee, via derived Nash Equilibria, that each participating agent's best strategy is to submit a model safer than a prescribed minimum-safety threshold. Empirical results show that our regulatory auction boosts safety and participation rates by 20% and 15% respectively, outperforming simple regulatory frameworks that merely enforce minimum safety standards.

LLMs are ideal for decision-making due to their ability to reason over long contexts and identify critical factors. However, challenges arise when processing transcripts of spoken speech describing complex scenarios. These transcripts often contain ungrammatical or incomplete sentences, repetitions, hedging, and vagueness. For example, during a company's earnings call, an executive might project a positive revenue outlook to reassure investors, despite significant uncertainty regarding future earnings. It is crucial for LLMs to incorporate this uncertainty systematically when making decisions. In this paper, we introduce DeFine, a new framework that constructs probabilistic factor profiles from complex scenarios. DeFine then integrates these profiles with analogical reasoning, leveraging insights from similar past experiences to guide LLMs in making critical decisions in novel situations. Our framework separates the tasks of quantifying uncertainty in complex scenarios and incorporating it into LLM decision-making. This approach is particularly useful in fields such as medical consultations, negotiations, and political debates, where making decisions under uncertainty is vital.

Recent retrieval-augmented models enhance basic methods by building a hierarchical structure over retrieved text chunks through recursive embedding, clustering, and summarization. The most relevant information is then retrieved from both the original text and generated summaries. However, such approaches face limitations with dynamic datasets, where adding or removing documents over time complicates the updating of hierarchical representations formed through clustering. We propose a new algorithm to efficiently maintain the recursive-abstractive tree structure in dynamic datasets, without compromising performance. Additionally, we introduce a novel post-retrieval method that applies query-focused recursive abstractive processing to substantially improve context quality. Our method overcomes the limitations of other approaches by functioning as a black-box post-retrieval layer compatible with any retrieval algorithm. Both algorithms are validated through extensive experiments on real-world datasets, demonstrating their effectiveness in handling dynamic data and improving retrieval performance.

Random Forests are widely recognized for establishing efficacy in classification and regression tasks, standing out in various domains such as medical diagnosis, finance, and personalized recommendations. These domains, however, are inherently sensitive to privacy concerns, as personal and confidential data are involved. With increasing demand for the right to be forgotten, particularly under regulations such as GDPR and CCPA, the ability to perform machine unlearning has become crucial for Random Forests. However, insufficient attention was paid to this topic, and existing approaches face difficulties in being applied to real-world scenarios. Addressing this gap, we propose the DynFrs framework designed to enable efficient machine unlearning in Random Forests while preserving predictive accuracy. Dynfrs leverages subsampling method Occ(q) and a lazy tag strategy Lzy, and is still adaptable to any Random Forest variant. In essence, Occ(q) ensures that each sample in the training set occurs only in a proportion of trees so that the impact of deleting samples is limited, and Lzy delays the reconstruction of a tree node until necessary, thereby avoiding unnecessary modifications on tree structures. In experiments, applying Dynfrs on Extremely Randomized Trees yields substantial improvements, achieving orders of magnitude faster unlearning performance and better predictive accuracy than existing machine unlearning methods for Random Forests.

Stable Diffusion Prompt: "A tall medieval knight in shining armor standing in a dense, dark forest, surrounded by towering trees with sunlight filtering through the leaves, the knight holding a sword and shield, an ancient castle visible in the distant mist, cinematic lighting." Example 3: Question: "How does a sunset over the ocean look?" Stable Diffusion Prompt: "A stunning sunset over the calm ocean, orange and pink skies reflecting on the water, waves gently rolling towards the shore, distant mountains on the horizon, seagulls flying overhead, a peaceful, serene atmosphere, watercolor painting style." Example 4: Question: "What does a futuristic robot look like in a factory?" Stable Diffusion Prompt: "A sleek, silver futuristic robot with glowing blue eyes working in a large industrial factory, surrounded by mechanical equipment, robotic arms assembling products, bright overhead lights casting shadows, high-tech machinery and cables everywhere, realistic style." Example 5: Question: "What does a peaceful village look like at night?" Stable Diffusion Prompt: "A quiet, peaceful village at night, small wooden houses with glowing windows,

Counterspeech presents a viable alternative to banning or suspending users for hate speech while upholding freedom of expression. However, writing effective counterspeech is challenging for moderators/users. Hence, developing suggestion tools for writing counterspeech is the need of the hour. One critical challenge in developing such a tool is the lack of quality and diversity of the responses in the existing datasets. Hence, we introduce a new dataset - CrowdCounter containing 3,425 hate speech-counterspeech pairs spanning six different counterspeech types (empathy, humor, questioning, warning, shaming, contradiction), which is the first of its kind. The design of our annotation platform itself encourages annotators to write type-specific, non-redundant and high-quality counterspeech. We evaluate two frameworks for generating counterspeech responses - vanilla and type-controlled prompts - across four large language models. In terms of metrics, we evaluate the responses using relevance, diversity and quality. We observe that Flan-T5 is the best model in the vanilla framework across different models. Type-specific prompts enhance the relevance of the responses, although they might reduce the language quality. DialoGPT proves to be the best at following the instructions and generating the type-specific counterspeech accurately.

The development and evaluation of Large Language Models (LLMs) have largely focused on individual capabilities. However, this overlooks the intersection of multiple abilities across different types of expertise that are often required for real-world tasks, which we term cross capabilities. To systematically explore this concept, we first define seven core individual capabilities and then pair them to form seven common cross capabilities, each supported by a manually constructed taxonomy. Building on these definitions, we introduce CrossEval, a benchmark comprising 1,400 human-annotated prompts, with 100 prompts for each individual and cross capability. To ensure reliable evaluation, we involve expert annotators to assess 4,200 model responses, gathering 8,400 human ratings with detailed explanations to serve as reference examples. Our findings reveal that, in both static evaluations and attempts to enhance specific abilities, current LLMs consistently exhibit the "Law of the Weakest Link," where cross-capability performance is significantly constrained by the weakest component. Specifically, across 58 cross-capability scores from 17 models, 38 scores are lower than all individual capabilities, while 20 fall between strong and weak, but closer to the weaker ability. These results highlight the under-performance of LLMs in cross-capability tasks, making the identification and improvement of the weakest capabilities a critical priority for future research to optimize performance in complex, multi-dimensional scenarios.