The increasing capabilities of large generative models and their ever more widespread deployment have raised concerns about their reliability, safety, and potential misuse. To address these issues, recent works have proposed to control model generation by steering model activations in order to effectively induce or prevent the emergence of concepts or behaviors in the generated output. In this paper we introduce Activation Transport (AcT), a general framework to steer activations guided by optimal transport theory that generalizes many previous activation-steering works. AcT is modality-agnostic and provides fine-grained control over the model behavior with negligible computational overhead, while minimally impacting model abilities. We experimentally show the effectiveness and versatility of our approach by addressing key challenges in large language models (LLMs) and text-to-image diffusion models (T2Is). For LLMs, we show that AcT can effectively mitigate toxicity, induce arbitrary concepts, and increase their truthfulness. In T2Is, we show how AcT enables fine-grained style control and concept negation.

Large Models (LMs) have heightened expectations for the potential of general AI as they are akin to human intelligence. This paper shows that recent large models such as Stable Diffusion and DALL-E3 also share the vulnerability of human intelligence, namely the "white bear phenomenon". We investigate the causes of the white bear phenomenon by analyzing their representation space. Based on this analysis, we propose a simple prompt-based attack method, which generates figures prohibited by the LM provider's policy. To counter these attacks, we introduce prompt-based defense strategies inspired by cognitive therapy techniques, successfully mitigating attacks by up to 48.22\%.

Existing methods for controlling language models, such as RLHF and Constitutional AI, involve determining which LLM behaviors are desirable and training them into a language model. However, in many cases, it is desirable for LLMs to be controllable \textit{at inference time}, so that they can be used in multiple contexts with diverse needs. We illustrate this with the \textbf{Pink Elephant Problem}: instructing an LLM to avoid discussing a certain entity (a ``Pink Elephant''), and instead discuss a preferred entity (``Grey Elephant''). We apply a novel simplification of Constitutional AI, \textbf{Direct Principle Feedback}, which skips the ranking of responses and uses DPO directly on critiques and revisions. Our results show that after DPF fine-tuning on our synthetic Pink Elephants dataset, our 13B fine-tuned LLaMA 2 model significantly outperforms Llama-2-13B-Chat and a prompted baseline, and performs as well as GPT-4 in on our curated test set assessing the Pink Elephant Problem.

A child who has never seen a pink elephant can still describe one -- unlike a computer. "The computer learns from data," says Jiajun Wu, a PhD student at MIT. "The ability to generalize and recognize something you've never seen before -- a pink elephant -- is very hard for machines." Deep learning systems interpret the world by picking out statistical patterns in data. This form of machine learning is now everywhere, automatically tagging friends on Facebook, narrating Alexa's latest weather forecast, and delivering fun facts via Google search. But statistical learning has its limits.

A child who has never seen a pink elephant can still describe one -- unlike a computer. "The computer learns from data," says Jiajun Wu, a PhD student at MIT. "The ability to generalize and recognize something you've never seen before -- a pink elephant -- is very hard for machines." Deep learning systems interpret the world by picking out statistical patterns in data. This form of machine learning is now everywhere, automatically tagging friends on Facebook, narrating Alexa's latest weather forecast, and delivering fun facts via Google search. But statistical learning has its limits.

If you want to teach a child what an elephant looks like, you have an infinite number of options. Take a photo from National Geographic, a stuffed animal of Dumbo, or an elephant keychain; show it to the child; and the next time he sees an object which looks like an elephant he will likely point and say the word. Teaching AI what an elephant looks like is a bit different. To train a machine learning algorithm, you will likely need thousands of elephant images using different perspectives, such as head, tail, and profile. But then, even after ingesting thousands of photos, if you connect your algorithm to a camera and show it a pink elephant keychain, it likely won't recognize it as an elephant.