arXiv subjects

Adam Kaufman

Publications and source records attributed to Adam Kaufman.

8 recordsLinked to original sources

arXiv2026

Retrying vs Resampling in AI Control

AI coding scaffolds like Claude Code and Codex use retrying: blocking actions flagged as risky and continuing the trajectory. We study retrying from an AI control perspective, which treats the model as potentially adversarial. We find that while retrying reduces honest suspicion scores, the untrusted model can exploit monitor rationale to construct sneakier attacks, negating safety gains. We also study resampling: drawing multiple samples from the same context, which does not leak exploitable information. We disentangle design choices that previous work on resampling had bundled together. In BashArena, with Claude Opus 4.6 as the untrusted model and MiMo-V2-Flash as the trusted monitor, drawing five samples per step and auditing on the maximum suspicion score raises safety from 61% to 71% at a 0.3% audit budget, at no cost to usefulness. Selectively resampling only the steps that look suspicious on the first draw recovers 6.2 percentage points of the gain while drawing only 10% as many extra samples. Two of our findings in this setting contradict earlier work on resampling. The first is that auditing based on the maximum across resampled suspicion scores outperforms using the minimum, which is the opposite of what Ctrl-Z found. The second is that executing the least suspicious sample, which is the central mechanism in earlier defer-to-resample protocols, gives only a small empirical safety gain in our setting (+3.9 pp, with the confidence interval overlapping zero).

cs.AI↗

arXiv2026

Realizing multi-orbital Emery models with ultracold atoms

Strongly-correlated electrons in transition-metal oxides give rise to intriguing emergent phenomena, including high-temperature superconductivity in cuprates. While simplified one-band Hubbard models capture some aspects, explicitly describing the interplay of copper and oxygen orbitals -- as in the three-band Emery model -- is essential to capture the full phenomenology of cuprates. Quantum simulators based on ultracold atoms offer a promising route to study such systems in a controlled setting, but realizing realistic multi-orbital Hubbard models remains challenging. Here we propose an optical superlattice architecture that implements the three-band Emery model with ultracold fermions. By combining lattice beams with controllable interference, we engineer orbital degrees of freedom that reproduce key features of the cuprate band structure, while enabling independent control of orbital-dependent interactions and charge-transfer energy. We show that single-particle quantum walks can benchmark the resulting tight-binding model. Using determinant quantum Monte Carlo, we further investigate thermodynamic properties in the undoped regime and find a finite-temperature metal-insulator crossover accompanied by the onset of antiferromagnetic correlations accessible in current experiments. Finally, we apply a Hamiltonian learning protocol enabling to infer effective single-band Hubbard models from experimental realizations of Emery models. Our results provide a practical pathway to simulate multi-orbital Hubbard physics with quantum gas microscopes.

cond-mat.quant-gas↗

arXiv2025

BashArena: A Control Setting for Highly Privileged AI Agents

Future AI agents might run autonomously with elevated privileges. If these agents are misaligned, they might abuse these privileges to cause serious damage. The field of AI control develops techniques that make it harder for misaligned AIs to cause such damage, while preserving their usefulness. We introduce BashArena, a setting for studying AI control techniques in security-critical environments. BashArena contains 637 Linux system administration and infrastructure engineering tasks in complex, realistic environments, along with four sabotage objectives (execute malware, exfiltrate secrets, escalate privileges, and disable firewall) for a red team to target. We evaluate multiple frontier LLMs on their ability to complete tasks, perform sabotage undetected, and detect sabotage attempts. Claude Sonnet 4.5 successfully executes sabotage while evading monitoring by GPT-4.1 mini 26% of the time, at 4% trajectory-wise FPR. Our findings provide a baseline for designing more effective control protocols in BashArena. We release the dataset as a ControlArena setting and share our task generation pipeline.

cs.CR↗

arXiv2025

Towards self-correcting quantum codes for neutral atom arrays

Discovering low-overhead quantum error-correcting codes is of significant interest for fault-tolerant quantum computation. For hardware capable of long-range connectivity, the bivariate bicycle codes offer significant overhead reduction compared to surface codes with similar performance. In this work, we present "ZSZ codes", a simple non-abelian generalization of the bivariate bicycle codes based on the group $\mathbb{Z}_\ell \rtimes \mathbb{Z}_m$. We numerically demonstrate that certain instances of this code family achieve competitive performance with the bivariate bicycle codes under circuit-level depolarizing noise using a belief-propagation and ordered-statistics decoder, with an observed threshold around $0.5\%$. We also benchmark the performance of this code family under local "self-correcting" decoders, where we observe significant improvements over the bivariate bicycle codes, including evidence of a sustainable threshold around $0.095\%$, which is higher than the $0.06\%$ that we estimate for the four-dimensional toric code under the same noise model. These results suggest that ZSZ codes are promising candidates for scalable self-correcting quantum memories. Finally, we describe how ZSZ codes can be realized with neutral atoms trapped in movable tweezer arrays, where a complete round of syndrome extraction can be achieved using simple global motions of the atomic arrays.

quant-ph↗

arXiv2025

Ctrl-Z: Controlling AI Agents via Resampling

Control evaluations measure whether monitoring and security protocols for AI systems prevent intentionally subversive AI models from causing harm. Our work presents the first control evaluation performed in an agent environment. We construct BashBench, a dataset of 257 challenging multi-step system administration tasks, and evaluate whether various safety measures can prevent an adversarially constructed AI agent from covertly downloading and executing malicious code in this environment. This multi-step setting introduces new attack and defense dynamics, which we investigate in order to design novel control protocols that prevent safety failures without hindering the ability of non-malicious agents to perform useful work. We introduce a class of control protocols called resample protocols that dynamically take additional samples of certain actions. We find these protocols significantly improve on existing techniques by selectively blocking the AI agent from executing suspicious code and incriminating the agent by generating additional examples of dangerous behavior. We measure the tradeoff between attack prevention and usefulness; our best protocol combines resampling with analysis of previous steps, reducing the success rate of attacks from 58% to 7% at a 5% cost to the performance of a non-malicious agent.

cs.LG↗

arXiv2022

Human Activity Recognition in an Open World

Managing novelty in perception-based human activity recognition (HAR) is critical in realistic settings to improve task performance over time and ensure solution generalization outside of prior seen samples. Novelty manifests in HAR as unseen samples, activities, objects, environments, and sensor changes, among other ways. Novelty may be task-relevant, such as a new class or new features, or task-irrelevant resulting in nuisance novelty, such as never before seen noise, blur, or distorted video recordings. To perform HAR optimally, algorithmic solutions must be tolerant to nuisance novelty, and learn over time in the face of novelty. This paper 1) formalizes the definition of novelty in HAR building upon the prior definition of novelty in classification tasks, 2) proposes an incremental open world learning (OWL) protocol and applies it to the Kinetics datasets to generate a new benchmark KOWL-718, 3) analyzes the performance of current state-of-the-art HAR models when novelty is introduced over time, 4) provides a containerized and packaged pipeline for reproducing the OWL protocol and for modifying for any future updates to Kinetics. The experimental analysis includes an ablation study of how the different models perform under various conditions as annotated by Kinetics-AVA. The protocol as an algorithm for reproducing experiments using the KOWL-718 benchmark will be publicly released with code and containers at https://github.com/prijatelj/human-activity-recognition-in-an-open-world. The code may be used to analyze different annotations and subsets of the Kinetics datasets in an incremental open world fashion, as well as be extended as further updates to Kinetics are released.

cs.CV↗

arXiv2020

Deblurring using Analysis-Synthesis Networks Pair

Blind image deblurring remains a challenging problem for modern artificial neural networks. Unlike other image restoration problems, deblurring networks fail behind the performance of existing deblurring algorithms in case of uniform and 3D blur models. This follows from the diverse and profound effect that the unknown blur-kernel has on the deblurring operator. We propose a new architecture which breaks the deblurring network into an analysis network which estimates the blur, and a synthesis network that uses this kernel to deblur the image. Unlike existing deblurring networks, this design allows us to explicitly incorporate the blur-kernel in the network's training. In addition, we introduce new cross-correlation layers that allow better blur estimations, as well as unique components that allow the estimate blur to control the action of the synthesis deblurring action. Evaluating the new approach over established benchmark datasets shows its ability to achieve state-of-the-art deblurring accuracy on various tests, as well as offer a major speedup in runtime.

cs.CV↗

arXiv2017

Realizing and Adiabatically Preparing Bosonic Integer and Fractional Quantum Hall states in Optical Lattices

We study the ground states of 2D lattice bosons in an artificial gauge field. Using state of the art DMRG simulations we obtain the zero temperature phase diagram for hardcore bosons at densities $n_b$ with flux $n_\phi$ per unit cell, which determines a filling $\nu=n_b/n_\phi$. We find several robust quantum Hall phases, including (i) a bosonic integer quantum Hall phase (BIQH) at $\nu=2$, that realizes an interacting symmetry protected topological phase in 2D (ii) bosonic fractional quantum Hall phases including robust states at $\nu=2/3$ and a Laughlin state at $\nu=1/2$. The observed states correspond to the bosonic Jain sequence ($\nu=p/(p+1)$) pointing towards an underlying composite fermion picture. In addition to identifying Hamiltonians whose ground states realize these phases, we discuss their preparation beginning from independent chains, and ramping up interchain couplings. Using time dependent DMRG simulations, these are shown to reliably produce states close to the ground state for experimentally relevant system sizes. Besides the wave-function overlap, we utilize a simple physical signature of these phases, the non-monotonic behavior of a two-point correlation, a direct consequence of edge states in a finite system, to numerically assess the effectiveness of the preparation scheme. Our proposal only utilizes existing experimental capabilities.

cond-mat.str-el↗