The mission of the Copilot Tuning Research and Applied Science teams is to ensure that AI can be easily adapted and tuned to enable critical, high-value applications, including entirely new productivity tools. We are creating algorithms, architectures, data, and eval pipelines to adapt LLMs, reasoning…
FIDO synced passkeys address account recovery challenges by enabling users to back up their FIDO2 private signing keys to the cloud storage of passkey management services (PMS). However, it introduces a serious security risk — attackers can steal users' passkeys through breaches of PMS's cloud…
Despite improvements in authentication mechanisms, compromise of online accounts remains prevalent. Therefore, technologies to detect compromise retroactively are also necessary. Service providers try to help users diagnose the security status of their accounts via account security interfaces (ASIs) that display recent logins or other activity.…
Avatars are edging into mainstream videoconferencing, but evaluation of how avatar animation modalities contribute to work meeting outcomes has been limited. We report a within-group videoconferencing experiment in which 68 employees of a global technology company, in 16 groups, used the same stylized avatars in…
Importance Multiple choice questions (MCQs) are an important and integral component of ophthalmology residency training evaluation and board certification; however, high-quality questions are difficult and time-consuming to draft. Objective To evaluate whether general-domain large language models (LLMs), particularly OpenAI’s Generative Pre-trained Transformer 4 (GPT-4), can reliably generate…
Recent advances in text-to-image models have increased the exposure of powerful image editing techniques as a tool, raising concerns about their potential for malicious use. An emerging line of research to address such threats focuses on implanting “protective” adversarial noise into images before their public…
While methods exist for aligning flow matching models -- a popular and effective class of generative models -- with human preferences, existing approaches fail to achieve both adaptation efficiency and probabilistically sound prior preservation. In this work, we leverage the theory of optimal control and…
We study the contextual dueling bandit problem, where a learner uses contextual information to make two decisions and receives only relative (comparison) feedback. This problem is crucial in reinforcement learning with human feedback (RLHF), widely applied in AI alignment to integrate human preferences into AI…
Designing functional proteins is a critical yet challenging problem due to the intricate interplay between backbone structures, sequences, and side-chains. Current approaches often decompose protein design into separate tasks, which can lead to accumulated errors, while recent efforts increasingly focus on all-atom protein design. However,…
Tests of conditional independence (CI) underpin a number of important problems in machine learning and statistics, from causal discovery to evaluation of predictor fairness and out-of-distribution robustness. Shah and Peters (2020) showed that, contrary to the unconditional case, no universally finite-sample valid test can ever…
The ability of LLM agents to plan and invoke tools exposes them to new safety risks, making a comprehensive red-teaming system crucial for discovering vulnerabilities and ensuring their safe deployment. We present SIRAJ: a generic red-teaming framework for arbitrary black-box LLM agents. We employ a…
Machine learning systems deployed in the wild must operate reliably despite unreliable inputs, whether arising from distribution shifts, adversarial manipulation, or strategic behavior by users. Content moderation is a prime example: violators deliberately exploit euphemisms, obfuscations, or benign co-occurrence patterns to evade detection, creating unreliable…
