Modern training strategies of deep neural networks (NNs) tend to induce a heavy-tailed (HT) spectra of layer weights. Extensive efforts to study this phenomenon have found that NNs with HT weight spectra tend to generalize well. A prevailing notion for the occurrence of such HT spectra attributes gradient noise during training as a key contributing factor. Our work shows that gradient noise is unnecessary for generating HT weight spectra: two-layer NNs trained with full-batch Gradient Descent/Adam can exhibit HT spectra in their weights after finite training steps. To this end, we first identify the scale of the learning rate at which one step of full-batch Adam can lead to feature learning in the shallow NN, particularly when learning a single index teacher model. Next, we show that multiple optimizer steps with such (sufficiently) large learning rates can transition the bulk of the weight's spectra into an HT distribution. To understand this behavior, we present a novel perspective based on the singular vectors of the weight matrices and optimizer updates. We show that the HT weight spectrum originates from the `spike', which is generated from feature learning and interacts with the main bulk to generate an HT spectrum. Finally, we analyze the correlations between the HT weight spectra and generalization after multiple optimizer updates with varying learning rates.

Large language models (LLMs) are being rapidly developed, and a key component of their widespread deployment is their safety-related alignment. Many red-teaming efforts aim to jailbreak LLMs, where among these efforts, the Greedy Coordinate Gradient (GCG) attack's success has led to a growing interest in the study of optimization-based jailbreaking techniques. Although GCG is a significant milestone, its attacking efficiency remains unsatisfactory. In this paper, we present several improved (empirical) techniques for optimization-based jailbreaks like GCG. We first observe that the single target template of "Sure" largely limits the attacking performance of GCG; given this, we propose to apply diverse target templates containing harmful self-suggestion and/or guidance to mislead LLMs. Besides, from the optimization aspects, we propose an automatic multi-coordinate updating strategy in GCG (i.e., adaptively deciding how many tokens to replace in each step) to accelerate convergence, as well as tricks like easy-to-hard initialisation. Then, we combine these improved technologies to develop an efficient jailbreak method, dubbed I-GCG. In our experiments, we evaluate on a series of benchmarks (such as NeurIPS 2023 Red Teaming Track). The results demonstrate that our improved techniques can help GCG outperform state-of-the-art jailbreaking attacks and achieve nearly 100% attack success rate. The code is released at https://github.com/jiaxiaojunQAQ/I-GCG.

Nevertheless, is it possible to use few-shot demonstrations to efficiently jailbreak LLMs within limited context sizes? While the vanilla few-shot jailbreaking may be inefficient, we propose improved techniques such as injecting special system tokens like [/INST] and employing demo-level random search from a collected demo pool. These simple techniques result in surprisingly effective jailbreaking against aligned LLMs (even with advanced defenses). For example, our method achieves > 80% (mostly > 95%) ASRs on Llama-2-7B and Llama-3-8B without multiple restarts, even if the models are enhanced by strong defenses such as perplexity detection and/or SmoothLLM, which is challenging for suffix-based jailbreaking. In addition, we conduct comprehensive and elaborate (e.g., making sure to use correct system prompts) evaluations against other aligned LLMs and advanced defenses, where our method consistently achieves nearly 100% ASRs. Our code is available at https://github.com/sail-sg/I-FSJ.

The surge in Large Language Models (LLMs) has revolutionized natural language processing, but fine-tuning them for specific tasks often encounters challenges in balancing performance and preserving general instruction-following abilities. In this paper, we posit that the distribution gap between task datasets and the LLMs serves as the primary underlying cause. To address the problem, we introduce Self-Distillation Fine-Tuning (SDFT), a novel approach that bridges the distribution gap by guiding fine-tuning with a distilled dataset generated by the model itself to match its original distribution. Experimental results on the Llama-2-chat model across various benchmarks demonstrate that SDFT effectively mitigates catastrophic forgetting while achieving comparable or superior performance on downstream tasks compared to the vanilla fine-tuning. Moreover, SDFT demonstrates the potential to maintain the helpfulness and safety alignment of LLMs. Our code is available at https://github.com/sail-sg/sdft.

The rapid adoption of text-to-image diffusion models in society underscores an urgent need to address their biases. Without interventions, these biases could propagate a skewed worldview and restrict opportunities for minority groups. In this work, we frame fairness as a distributional alignment problem. Our solution consists of two main technical contributions: (1) a distributional alignment loss that steers specific characteristics of the generated images towards a user-defined target distribution, and (2) adjusted direct finetuning of diffusion model's sampling process (adjusted DFT), which leverages an adjusted gradient to directly optimize losses defined on the generated images. Empirically, our method markedly reduces gender, racial, and their intersectional biases for occupational prompts. Gender bias is significantly reduced even when finetuning just five soft tokens. Crucially, our method supports diverse perspectives of fairness beyond absolute equality, which is demonstrated by controlling age to a $75\%$ young and $25\%$ old distribution while simultaneously debiasing gender and race. Finally, our method is scalable: it can debias multiple concepts at once by simply including these prompts in the finetuning data. We share code and various fair diffusion model adaptors at https://sail-sg.github.io/finetune-fair-diffusion/.

Recent research has made significant progress in optimizing diffusion models for specific downstream objectives, which is an important pursuit in fields such as graph generation for drug design. However, directly applying these models to graph diffusion presents challenges, resulting in suboptimal performance. This paper introduces graph diffusion policy optimization (GDPO), a novel approach to optimize graph diffusion models for arbitrary (e.g., non-differentiable) objectives using reinforcement learning. GDPO is based on an eager policy gradient tailored for graph diffusion models, developed through meticulous analysis and promising improved performance. Experimental results show that GDPO achieves state-of-the-art performance in various graph generation tasks with complex and diverse objectives. Code is available at https://github.com/sail-sg/GDPO.

The emerging success of large language models (LLMs) heavily relies on collecting abundant training data from external (untrusted) sources. Despite substantial efforts devoted to data cleaning and curation, well-constructed LLMs have been reported to suffer from copyright infringement, data poisoning, and/or privacy violations, which would impede practical deployment of LLMs. In this study, we propose a simple and easily implementable method for purifying LLMs from the negative effects caused by uncurated data, namely, through ensembling LLMs with benign and small language models (SLMs). Aside from theoretical guarantees, we perform comprehensive experiments to empirically confirm the efficacy of ensembling LLMs with SLMs, which can effectively preserve the performance of LLMs while mitigating issues such as copyright infringement, data poisoning, and privacy violations.

Backdoor attacks are commonly executed by contaminating training data, such that a trigger can activate predetermined harmful effects during the test phase. In this work, we present AnyDoor, a test-time backdoor attack against multimodal large language models (MLLMs), which involves injecting the backdoor into the textual modality using adversarial test images (sharing the same universal perturbation), without requiring access to or modification of the training data. AnyDoor employs similar techniques used in universal adversarial attacks, but distinguishes itself by its ability to decouple the timing of setup and activation of harmful effects. In our experiments, we validate the effectiveness of AnyDoor against popular MLLMs such as LLaVA-1.5, MiniGPT-4, InstructBLIP, and BLIP-2, as well as provide comprehensive ablation studies. Notably, because the backdoor is injected by a universal perturbation, AnyDoor can dynamically change its backdoor trigger prompts/harmful effects, exposing a new challenge for defending against backdoor attacks. Our project page is available at https://sail-sg.github.io/AnyDoor/.

A multimodal large language model (MLLM) agent can receive instructions, capture images, retrieve histories from memory, and decide which tools to use. Nonetheless, red-teaming efforts have revealed that adversarial images/prompts can jailbreak an MLLM and cause unaligned behaviors. In this work, we report an even more severe safety issue in multi-agent environments, referred to as infectious jailbreak. It entails the adversary simply jailbreaking a single agent, and without any further intervention from the adversary, (almost) all agents will become infected exponentially fast and exhibit harmful behaviors. To validate the feasibility of infectious jailbreak, we simulate multi-agent environments containing up to one million LLaVA-1.5 agents, and employ randomized pair-wise chat as a proof-of-concept instantiation for multi-agent interaction. Our results show that feeding an (infectious) adversarial image into the memory of any randomly chosen agent is sufficient to achieve infectious jailbreak. Finally, we derive a simple principle for determining whether a defense mechanism can provably restrain the spread of infectious jailbreak, but how to design a practical defense that meets this principle remains an open question to investigate. Our project page is available at https://sail-sg.github.io/Agent-Smith/.

Large language models (LLMs) are vulnerable to jailbreak attacks - resulting in harmful, unethical, or biased text generations. However, existing jailbreaking methods are computationally costly. In this paper, we propose the weak-to-strong jailbreaking attack, an efficient method to attack aligned LLMs to produce harmful text. Our key intuition is based on the observation that jailbroken and aligned models only differ in their initial decoding distributions. The weak-to-strong attack's key technical insight is using two smaller models (a safe and an unsafe one) to adversarially modify a significantly larger safe model's decoding probabilities. We evaluate the weak-to-strong attack on 5 diverse LLMs from 3 organizations. The results show our method can increase the misalignment rate to over 99% on two datasets with just one forward pass per example. Our study exposes an urgent safety issue that needs to be addressed when aligning LLMs. As an initial attempt, we propose a defense strategy to protect against such attacks, but creating more advanced defenses remains challenging. The code for replicating the method is available at https://github.com/XuandongZhao/weak-to-strong