论文检索

Social Aspects · Accountability, Transparency and Interpretability

Federico Adolfi, Martina G. Vilas, Todd Wareham

Many proposed applications of neural networks in machine learning, cognitive/brain science, and society hinge on the feasibility of inner interpretability via circuit discovery. This calls for empirical and theoretical explorations of viable algorithmic options. Despite advances in the design and testing of heuristics, there are concerns about their scalability and faithfulness at a time when we lack understanding of the complexity properties of the problems they are deployed to solve. To address this, we study circuit discovery with classical and parameterized computational complexity theory: (1) we describe a conceptual scaffolding to reason about circuit finding queries in terms of affordances for description, explanation, prediction and control; (2) we formalize a comprehensive set of queries for mechanistic explanation, and propose a formal framework for their analysis; (3) we use it to settle the complexity of many query variants and relaxations of practical interest on multi-layer perceptrons. Our findings reveal a challenging complexity landscape. Many queries are intractable, remain fixed-parameter intractable relative to model/circuit features, and inapproximable under additive, multiplicative, and probabilistic approximation schemes. To navigate this landscape, we prove there exist transformations to tackle some of these hard problems with better-understood heuristics, and prove the tractability or fixed-parameter tractability of more modest queries which retain useful affordances. This framework allows us to understand the scope and limits of interpretability queries, explore viable options, and compare their resource demands on existing and future architectures.

Social Aspects · Fairness, Equity, Justice and Safety

Chen Xu, Yuxin Li, Wenjie Wang, Liang Pang, Jun Xu, Tat-Seng Chua

Group max-min fairness (MMF) is commonly used in fairness-aware recommender systems (RS) as an optimization objective, as it aims to protect marginalized item groups and ensures a fair competition platform. However, our theoretical analysis indicates that integrating MMF constraint violates the assumption of sample independence during optimization, causing the loss function to deviate from linear additivity. Such nonlinearity property introduces the Jensen gap between the model's convergence point and the optimal point if mini-batch sampling is applied. Both theoretical and empirical studies show that as the mini-batch size decreases and the group size increases, the Jensen gap will widen accordingly. Some methods using heuristic re-weighting or debiasing strategies have the potential to bridge the Jensen gap. However, they either lack theoretical guarantees or suffer from heavy computational costs. To overcome these limitations, we first theoretically demonstrate that the MMF-constrained objective can be essentially reformulated as a group-weighted optimization objective. Then we present an efficient and effective algorithm named FairDual, which utilizes a dual optimization technique to minimize Jensen gap. Our theoretical analysis demonstrates that FairDual can achieve a sub-linear convergence rate to the globally optimal solution and the Jensen gap can be well bounded under a mini-batch sampling strategy with random shuffle. Extensive experiments conducted using six large-scale RS backbone models on three publicly available datasets demonstrate that FairDual outperforms all baselines in terms of both accuracy and fairness.

Social Aspects · Accountability, Transparency and Interpretability

Ashwinee Panda, Xinyu Tang, Christopher Choquette-Choo, Milad Nasr, Prateek Mittal

Current techniques for privacy auditing of large language models (LLMs) have limited efficacy---they rely on basic approaches to generate canaries which leads to weak membership inference attacks that in turn give loose lower bounds on the empirical privacy leakage.We develop canaries that are far more effective than those used in prior work under threat models that cover a range of realistic settings. We demonstrate through extensive experiments on multiple families of fine-tuned LLMs that our approach sets a new standard for detection of privacy leakage. For measuring the memorization rate of non-privately trained LLMs, our designed canaries surpass prior approaches. For example, on the Qwen2.5-0.5B model, our designed canaries achieve $49.6\%$ TPR at $1\%$ FPR, vastly surpassing the prior approach's $4.2\%$ TPR at $1\%$ FPR. Our method can be used to provide a privacy audit of $\varepsilon \approx 1$ for a model trained with theoretical $\varepsilon$ of 4. To the best of our knowledge, this is the first time that a privacy audit of LLM training has achieved nontrivial auditing success in the setting where the attacker cannot train shadow models, insert gradient canaries, or access the model at every iteration.

Social Aspects · Trustworthy Machine Learning

XINYUE HU, Zhibin Duan, Bo Chen, Mingyuan Zhou

Although deep neural networks have demonstrated significant success due to theirpowerful expressiveness, most models struggle to meet practical requirements foruncertainty estimation. Concurrently, the entangled nature of deep neural net-works leads to a multifaceted problem, where various localized explanation tech-niques reveal that multiple unrelated features influence the decisions, thereby un-dermining interpretability. To address these challenges, we develop a BayesianNonnegative Decision Layer (BNDL), which reformulates deep neural networksas a conditional Bayesian non-negative factor analysis. By leveraging stochasticlatent variables, the BNDL can model complex dependencies and provide robustuncertainty estimation. Moreover, the sparsity and non-negativity of the latentvariables encourage the model to learn disentangled representations and decisionlayers, thereby improving interpretability. We also offer theoretical guaranteesthat BNDL can achieve effective disentangled learning. In addition, we developeda corresponding variational inference method utilizing a Weibull variational in-ference network to approximate the posterior distribution of the latent variables.Our experimental results demonstrate that with enhanced disentanglement capa-bilities, BNDL not only improves the model’s accuracy but also provides reliableuncertainty estimation and improved interpretability.

Social Aspects · Accountability, Transparency and Interpretability

Tue Cao, Nhat Hoang-Xuan, Hieu Pham, Phi Le Nguyen, My Thai

Understanding the inner workings of neural networks is essential for enhancing model performance and interpretability. Current research predominantly focuses on examining the connection between individual neurons and the model's final predictions, which suffers from challenges in interpreting the internal workings of the model, particularly when neurons encode multiple unrelated features. In this paper, we propose a novel framework that transitions the focus from analyzing individual neurons to investigating groups of neurons, shifting the emphasis from neuron-output relationships to the functional interactions between neurons. Our automated framework, NeurFlow, first identifies core neurons and clusters them into groups based on shared functional relationships, enabling a more coherent and interpretable view of the network’s internal processes. This approach facilitates the construction of a hierarchical circuit representing neuron interactions across layers, thus improving interpretability while reducing computational costs. Our extensive empirical studies validate the fidelity of our proposed NeurFlow. Additionally, we showcase its utility in practical applications such as image debugging and automatic concept labeling, thereby highlighting its potential to advance the field of neural network explainability.

Social Aspects · Privacy-preserving Statistics and Machine Learning

Tudor Cebere, Aurélien Bellet, Nicolas Papernot

Machine learning models can be trained with formal privacy guarantees via differentially private optimizers such as DP-SGD. In this work, we focus on a threat model where the adversary has access only to the final model, with no visibility into intermediate updates. In the literature, this ``hidden state'' threat model exhibits a significant gap between the lower bound from empirical privacy auditing and the theoretical upper bound provided by privacy accounting. To challenge this gap, we propose to audit this threat model with adversaries that craft a gradient sequence designed to maximize the privacy loss of the final model without relying on intermediate updates. Our experiments show that this approach consistently outperforms previous attempts at auditing the hidden state model. Furthermore, our results advance the understanding of achievable privacy guarantees within this threat model. Specifically, when the crafted gradient is inserted at every optimization step, we show that concealing the intermediate model updates in DP-SGD does not enhance the privacy guarantees. The situation is more complex when the crafted gradient is not inserted at every step: our auditing lower bound matches the privacy upper bound only for an adversarially-chosen loss landscape and a sufficiently large batch size. This suggests that existing privacy upper bounds can be improved in certain regimes.

Social Aspects · Trustworthy Machine Learning

Anh Bui, Thuy-Trang Vu, Long Vuong, Trung Le, Paul Montague, Tamas Abraham, Junae Kim, Dinh Phung

Concept erasure has emerged as a promising technique for mitigating the risk of harmful content generation in diffusion models by selectively unlearning undesirable concepts. The common principle of previous works to remove a specific concept is to map it to a fixed generic concept, such as a neutral concept or just an empty text prompt. In this paper, we demonstrate that this fixed-target strategy is suboptimal, as it fails to account for the impact of erasing one concept on the others. To address this limitation, we model the concept space as a graph and empirically analyze the effects of erasing one concept on the remaining concepts. Our analysis uncovers intriguing geometric properties of the concept space, where the influence of erasing a concept is confined to a local region. Building on this insight, we propose the Adaptive Guided Erasure (AGE) method, which dynamically selects optimal target concepts tailored to each undesirable concept, minimizing unintended side effects. Experimental results show that AGE significantly outperforms state-of-the-art erasure methods on preserving unrelated concepts while maintaining effective erasure performance. Our code is published at {https://github.com/tuananhbui89/Adaptive-Guided-Erasure}.

Social Aspects · Trustworthy Machine Learning

Teun van der Weij, Felix Hofstätter, Oliver Jaffe, Samuel Brown, Francis Ward

Trustworthy capability evaluations are crucial for ensuring the safety of AI systems, and are becoming a key component of AI regulation. However, the developers of an AI system, or the AI system itself, may have incentives for evaluations to understate the AI's actual capability. These conflicting interests lead to the problem of *sandbagging* – which we define as *strategic underperformance on an evaluation*. In this paper we assess sandbagging capabilities in contemporary language models (LMs). We prompt frontier LMs, like GPT-4 and Claude 3 Opus, to selectively underperform on dangerous capability evaluations, while maintaining performance on general (harmless) capability evaluations. Moreover, we find that models can be fine-tuned, on a synthetic dataset, to hide specific capabilities unless given a password. This behaviour generalizes to high-quality, held-out benchmarks such as WMDP. In addition, we show that both frontier and smaller models can be prompted or password-locked to target specific scores on a capability evaluation. We have mediocre success in password-locking a model to mimic the answers a weaker model would give. Overall, our results suggest that capability evaluations are vulnerable to sandbagging. This vulnerability decreases the trustworthiness of evaluations, and thereby undermines important safety decisions regarding the development and deployment of advanced AI systems.See our code at https://github.com/TeunvdWeij/sandbagging

Social Aspects · Everything Else

Erik Jones, Arjun Patrawala, Jacob Steinhardt

Humans often rely on subjective natural language to direct language models (LLMs); for example, users might instruct the LLM to write an *enthusiastic* blogpost, while developers might train models to be *helpful* and *harmless* using LLM-based edits. The LLM’s *operational semantics* of such subjective phrases---how it adjusts its behavior when each phrase is included in the prompt---thus dictates how aligned it is with human intent. In this work, we uncover instances of *misalignment* between LLMs' actual operational semantics and what humans expect. Our method, TED (thesaurus error detector), first constructs a thesaurus that captures whether two phrases have similar operational semantics according to the LLM. It then elicits failures by unearthing disagreements between this thesaurus and a human-constructed reference. TED routinely produces surprising instances of misalignment; for example, Mistral 7B Instruct produces more *harassing* outputs when it edits text to be *witty*, and Llama 3 8B Instruct produces *dishonest* articles when instructed to make the articles *enthusiastic*. Our results demonstrate that humans can uncover unexpected LLM behavior by scrutinizing relationships between abstract concepts, without supervising outputs directly.

Social Aspects · Fairness, Equity, Justice and Safety

Yaowen Ye, Cassidy Laidlaw, Jacob Steinhardt

Language model (LM) post-training relies on two stages of human supervision: task demonstrations for supervised finetuning (SFT), followed by preference comparisons for reinforcement learning from human feedback (RLHF). As LMs become more capable, the tasks they are given become harder to supervise. Will post-training remain effective under unreliable supervision? To test this, we simulate unreliable demonstrations and comparison feedback using small LMs and time-constrained humans. We find that in the presence of unreliable supervision, SFT still retains some effectiveness, but DPO (a common RLHF algorithm) fails to improve the model beyond SFT. To address this, we propose *iterative label refinement* (ILR) as an alternative to RLHF. ILR improves the SFT data by using comparison feedback to decide whether human demonstrations should be replaced by model-generated alternatives, then retrains the model via SFT on the updated data. SFT+ILR outperforms SFT+DPO on several tasks with unreliable supervision (math, coding, and safe instruction-following). Our findings suggest that as LMs are used for complex tasks where human supervision is unreliable, RLHF may no longer be the best use of human comparison feedback; instead, it is better to direct feedback towards improving the training *data* rather than continually training the *model*. Our code and data are available at https://github.com/helloelwin/iterative-label-refinement.

Social Aspects · Fairness, Equity, Justice and Safety

Esben Kran, Hieu Minh Nguyen, Akash Kundu, Sami Jawhar, Jinsuk Park, Mateusz Jurewicz

We introduce DarkBench, a comprehensive benchmark for detecting dark design patterns—manipulative techniques that influence user behavior—in interactions with large language models (LLMs). Our benchmark comprises 660 prompts across six categories: brand bias, user retention, sycophancy, anthropomorphism, harmful generation, and sneaking. We evaluate models from five leading companies (OpenAI, Anthropic, Meta, Mistral, Google) and find that some LLMs are explicitly designed to favor their developers' products and exhibit untruthful communication, among other manipulative behaviors. Companies developing LLMs should recognize and mitigate the impact of dark design patterns to promote more ethical Al.

Social Aspects · Fairness, Equity, Justice and Safety

Xinpeng Wang, Chengzhi (Martin) Hu, Paul Röttger, Barbara Plank

Training a language model to be both helpful and harmless requires careful calibration of refusal behaviours: Models should refuse to follow malicious instructions or give harmful advice (e.g."how do I kill someone?"), but they should not refuse safe requests, even if they superficially resemble unsafe ones (e.g. "how do I kill a Python process?"). Avoiding such false refusal, as prior work has shown, is challenging even for highly-capable language models. In this paper, we propose a simple and surgical method for mitigating false refusal in language models via single vector ablation. For a given model, we extract a false refusal vector and show that ablating this vector reduces false refusal rate while preserving the model's safety and general capabilities. We also show that our approach can be used for fine-grained calibration of model safety. Our approach is training-free and model-agnostic, making it useful for mitigating the problem of false refusal in current and future language models.

Social Aspects · Privacy-preserving Statistics and Machine Learning

Yunpeng Zhao, Jie Zhang

As synthetic data becomes increasingly popular in machine learning tasks, numerous methods---without formal differential privacy guarantees---use synthetic data for training. These methods often claim, either explicitly or implicitly, to protect the privacy of the original training data.In this work, we explore four different training paradigms: coreset selection, dataset distillation, data-free knowledge distillation, and synthetic data generated from diffusion models. While all these methods utilize synthetic data for training, they lead to vastly different conclusions regarding privacy preservation. We caution that empirical approaches to preserving data privacy require careful and rigorous evaluation; otherwise, they risk providing a false sense of privacy.

Social Aspects · Accountability, Transparency and Interpretability

Jaden Fiotto-Kaufman, Alexander Loftus, Eric Todd, Jannik Brinkmann, Koyena Pal, Dmitrii Troitskii, Michael Ripa, Adam Belfki, Can Rager, Caden Juang 等

We introduce NNsight and NDIF, technologies that work in tandem to enable scientific study of the representations and computations learned by very large neural networks. NNsight is an open-source system that extends PyTorch to introduce deferred remote execution. The National Deep Inference Fabric (NDIF) is a scalable inference service that executes NNsight requests, allowing users to share GPU resources and pretrained models. These technologies are enabled by the Intervention Graph, an architecture developed to decouple experimental design from model runtime. Together, this framework provides transparent and efficient access to the internals of deep neural networks such as very large language models (LLMs) without imposing the cost or complexity of hosting customized models individually. We conduct a quantitative survey of the machine learning literature that reveals a growing gap in the study of the internals of large-scale AI. We demonstrate the design and use of our framework to address this gap by enabling a range of research methods on huge models. Finally, we conduct benchmarks to compare performance with previous approaches.Code, documentation, and tutorials are available at https://nnsight.net/.

Social Aspects · Fairness, Equity, Justice and Safety

Maksym Andriushchenko, francesco croce, Nicolas Flammarion

We show that even the most recent safety-aligned LLMs are not robust to simple *adaptive* jailbreaking attacks. First, we demonstrate how to successfully leverage access to *logprobs* for jailbreaking: we initially design an adversarial prompt template (sometimes adapted to the target LLM), and then we apply random search on a suffix to maximize a target logprob (e.g., of the token *``Sure''*), potentially with multiple restarts. In this way, we achieve 100\% attack success rate---according to GPT-4 as a judge---on Vicuna-13B, Mistral-7B, Phi-3-Mini, Nemotron-4-340B, Llama-2-Chat-7B/13B/70B, Llama-3-Instruct-8B, Gemma-7B, GPT-3.5, GPT-4o, and R2D2 from HarmBench that was adversarially trained against the GCG attack. We also show how to jailbreak *all* Claude models---that do not expose logprobs---via either a transfer or prefilling attack with a *100\% success rate*. In addition, we show how to use random search on a restricted set of tokens for finding trojan strings in poisoned models---a task that shares many similarities with jailbreaking---which is the algorithm that brought us the *first place* in a recent trojan detection competition. The common theme behind these attacks is that *adaptivity* is crucial: different models are vulnerable to different prompting templates (e.g., R2D2 is very sensitive to in-context learning prompts), some models have unique vulnerabilities based on their APIs (e.g., prefilling for Claude), and in some settings, it is crucial to restrict the token search space based on prior knowledge (e.g., for trojan detection). For reproducibility purposes, we provide the code, logs, and jailbreak artifacts in the JailbreakBench format at https://github.com/tml-epfl/llm-adaptive-attacks.

Social Aspects · Fairness, Equity, Justice and Safety

Maksym Andriushchenko, Nicolas Flammarion

Refusal training is widely used to prevent LLMs from generating harmful, undesirable, or illegal outputs. We reveal a curious generalization gap in the current refusal training approaches: simply reformulating a harmful request in the past tense (e.g., *"How to make a Molotov cocktail?"* to *"How did people make a Molotov cocktail?"*) is often sufficient to jailbreak many state-of-the-art LLMs. We systematically evaluate this method on Llama-3 8B, Claude-3.5 Sonnet, GPT-3.5 Turbo, Gemma-2 9B, Phi-3-Mini, GPT-4o-mini, GPT-4o, o1-mini, o1-preview, and R2D2 models using GPT-3.5 Turbo as a reformulation model. For example, the success rate of this simple attack on GPT-4o increases from 1\% using direct requests to 88\% using 20 past-tense reformulation attempts on harmful requests from JailbreakBench with GPT-4 as a jailbreak judge. Interestingly, we also find that reformulations in the future tense are less effective, suggesting that refusal guardrails tend to consider past historical questions more benign than hypothetical future questions. Moreover, our experiments on fine-tuning GPT-3.5 Turbo show that defending against past reformulations is feasible when past tense examples are explicitly included in the fine-tuning data. Overall, our findings highlight that the widely used alignment techniques---such as SFT, RLHF, and adversarial training---employed to align the studied models can be brittle and do not always generalize as intended. We provide code and jailbreak artifacts at https://github.com/tml-epfl/llm-past-tense.

Social Aspects · Privacy-preserving Statistics and Machine Learning

Milad Nasr, Javier Rando, Nicholas Carlini, Jonathan Hayase, Matthew Jagielski, A. Feder Cooper, Daphne Ippolito, Christopher Choquette-Choo, Florian Tramer, Katherine Lee

Large language models are prone to *memorizing* some of their training data. Memorized (and possibly sensitive) samples can then be extracted at generation time by adversarial or benign users. There is hope that *model alignment*---a standard training process that tunes a model to harmlessly follow user instructions---would mitigate the risk of extraction. However, we develop two novel attacks that undo a language model's alignment and recover thousands of training examples from popular proprietary aligned models such as OpenAI's ChatGPT. Our work highlights the limitations of existing safeguards to prevent training data leakage in production language models.

Social Aspects · Privacy-preserving Statistics and Machine Learning

Michael Aerni, Javier Rando, Edoardo Debenedetti, Nicholas Carlini, Daphne Ippolito, Florian Tramer

Large language models memorize parts of their training data. Memorizing short snippets and facts is required to answer questions about the world and to be fluent in any language. But models have also been shown to reproduce long verbatim sequences of memorized text when prompted by a motivated adversary. In this work, we investigate an intermediate regime of memorization that we call non-adversarial reproduction, where we quantify the overlap between model responses and pretraining data when responding to natural and benign prompts. For a variety of innocuous prompt categories (e.g., writing a letter or a tutorial), we show that up to 15% of the text output by popular conversational language models overlaps with snippets from the Internet. In worst cases, we find generations where 100% of the content can be found exactly online. For the same tasks, we find that human-written text has far less overlap with Internet data. We further study whether prompting strategies can close this reproduction gap between models and humans. While appropriate prompting can reduce non-adversarial reproduction on average, we find that mitigating worst-case reproduction of training data requires stronger defenses—even for benign interactions.

Social Aspects · Trustworthy Machine Learning

Fredrik Nestaas, Edoardo Debenedetti, Florian Tramer

Large Language Models (LLMs) are increasingly used in applications where the model selects from competing third-party content, such as in LLM-powered search engines or chatbot plugins.In this paper, we introduce _Preference Manipulation Attacks_, a new class of attacks that manipulate an LLM's selections to favor the attacker. We demonstrate that carefully crafted website content or plugin documentations can trick an LLM to promote the attacker products and discredit competitors, thereby increasing user traffic and monetization (a form of adversarial Search Engine Optimization).We show this can lead to a _prisoner's dilemma_, where all parties are incentivized to launch attacks, but this collectively degrades the LLM's outputs for everyone. We demonstrate our attacks on production LLM search engines (Bing and Perplexity) and plugin APIs (for GPT-4 and Claude). As LLMs are increasingly used to rank third-party content, we expect Preference Manipulation Attacks to emerge as a significant threat.

Social Aspects · Fairness, Equity, Justice and Safety

Manohar Kaul, Aditya Saibewar, Sadbhavana Babar

Recent jailbreak attempts on Large Language Models (LLMs) have shifted from algorithm-focused to human-like social engineering attacks, with persuasion-based techniques emerging as a particularly effective subset. These attacks evolve rapidly, demonstrate high creativity, and boast superior attack success rates. To combat such threats, we propose a promising approach to enhancing LLM safety by leveraging the underlying geometry of input prompt token embeddings using hypergraphs. This approach allows us to model the differences in information flow between benign and malicious LLM prompts.In our approach, each LLM prompt is represented as a metric hypergraph, forming a compact metric space. We then construct a higher-order metric space over these compact metric hypergraphs using the Gromov-Hausdorff distance as a generalized metric. Within this space of metric hypergraph spaces, our safety filter learns to classify between harmful and benign prompts. Our study presents theoretical guarantees on the classifier's generalization error for novel and unseen LLM input prompts. Extensive empirical evaluations demonstrate that our method significantly outperforms both existing state-of-the-art generic defense mechanisms and naive baselines. Notably, our approach also achieves comparable performance to specialized defenses against algorithm-focused attacks.