Tony T. Wang

Tony T. Wang, Igor Zablotchi, Nir Shavit et al.

We study the data-scaling of transfer learning from foundation models in the low-downstream-data regime. We observe an intriguing phenomenon which we call cliff-learning. Cliff-learning refers to regions of data-scaling laws where performance improves at a faster than power law rate (i.e. regions of concavity on a log-log scaling plot). We conduct an in-depth investigation of foundation-model cliff-learning and study toy models of the phenomenon. We observe that the degree of cliff-learning reflects the degree of compatibility between the priors of a learning algorithm and the task being learned.

Tony T. Wang, John Hughes, Henry Sleight et al.

Defending large language models against jailbreaks so that they never engage in a broadly-defined set of forbidden behaviors is an open problem. In this paper, we investigate the difficulty of jailbreak-defense when we only want to forbid a narrowly-defined set of behaviors. As a case study, we focus on preventing an LLM from helping a user make a bomb. We find that popular defenses such as safety training, adversarial training, and input/output classifiers are unable to fully solve this problem. In pursuit of a better solution, we develop a transcript-classifier defense which outperforms the baseline defenses we test. However, our classifier defense still fails in some circumstances, which highlights the difficulty of jailbreak-defense even in a narrow domain.

Tony T. Wang, Miles Wang, Kaivalya Hariharan et al.

LLMs often face competing pressures (for example helpfulness vs. harmlessness). To understand how models resolve such conflicts, we study Llama-2-chat models on the forbidden fact task. Specifically, we instruct Llama-2 to truthfully complete a factual recall statement while forbidding it from saying the correct answer. This often makes the model give incorrect answers. We decompose Llama-2 into 1000+ components, and rank each one with respect to how useful it is for forbidding the correct answer. We find that in aggregate, around 35 components are enough to reliably implement the full suppression behavior. However, these components are fairly heterogeneous and many operate using faulty heuristics. We discover that one of these heuristics can be exploited via a manually designed adversarial attack which we call The California Attack. Our results highlight some roadblocks standing in the way of being able to successfully interpret advanced ML systems. Project website available at https://forbiddenfacts.github.io .

Avichal Goel, Yoon Kim, Nir Shavit et al.

Finetuning (pretrained) language models is a standard approach for updating their internal parametric knowledge and specializing them to new tasks and domains. However, the corresponding model weight changes ("weight diffs") are not generally interpretable. While inspecting the finetuning dataset can give a sense of how the model might have changed, these datasets are often not publicly available or are too large to work with directly. Towards the goal of comprehensively understanding weight diffs in natural language, we introduce Diff Interpretation Tuning (DIT), a method that trains models to describe their own finetuning-induced modifications. Our approach uses synthetic, labeled weight diffs to train a DIT-adapter, which can be applied to a compatible finetuned model to make it describe how it has changed. We demonstrate in two proof-of-concept settings (reporting hidden behaviors and summarizing finetuned knowledge) that our method enables models to describe their finetuning-induced modifications using accurate natural language descriptions.

Blerim Abdullai, Tony Wang, Xinyuan Qiao et al.

GNSS is unreliable, inaccurate, and insufficient in many real-time autonomous field applications. In this work, we present a GNSS-free global localization solution that contains a method of registering imaging radar on the ground with overhead RGB imagery, with joint optimization of relative poses from odometry and global poses from our overhead registration. Previous works have used various combinations of ground sensors and overhead imagery, and different feature extraction and matching methods. These include various handcrafted and deep-learning-based methods for extracting features from overhead imagery. Our work presents insights on extracting essential features from RGB overhead images for effective global localization against overhead imagery using only ground radar and a single georeferenced initial guess. We motivate our method by evaluating it on datasets in diverse geographic conditions and robotic platforms, including on an Unmanned Surface Vessel (USV) as well as urban and suburban driving datasets.

Danny Halawi, Alexander Wei, Eric Wallace et al.

Black-box finetuning is an emerging interface for adapting state-of-the-art language models to user needs. However, such access may also let malicious actors undermine model safety. To demonstrate the challenge of defending finetuning interfaces, we introduce covert malicious finetuning, a method to compromise model safety via finetuning while evading detection. Our method constructs a malicious dataset where every individual datapoint appears innocuous, but finetuning on the dataset teaches the model to respond to encoded harmful requests with encoded harmful responses. Applied to GPT-4, our method produces a finetuned model that acts on harmful instructions 99% of the time and avoids detection by defense mechanisms such as dataset inspection, safety evaluations, and input/output classifiers. Our findings question whether black-box finetuning access can be secured against sophisticated adversaries.