Search for: All records

Safety is critical to the usage of large language models (LLMs). Multiple techniques such as data filtering and supervised fine tuning have been developed to strengthen LLM safety. However, currently known techniques presume that corpora used for safety alignment of LLMs are solely interpreted by semantics. This assumption, however, does not hold in real-world applications, which leads to severe vulnerabilities in LLMs. For example, users of forums often use ASCII art, a form of text-based art, to convey image information. In this paper, we propose a novel ASCII art-based jailbreak attack and introduce a comprehensive benchmark Vision-in-Text Challenge (VITC) to evaluate the capabilities of LLMs in recognizing prompts that cannot be solely interpreted by semantics. We show that five SOTA LLMs (GPT-3.5, GPT-4, Gemini, Claude, and Llama2) struggle to recognize prompts provided in the form of ASCII art. Based on this observation, we develop the jailbreak attack ArtPrompt, which leverages the poor performance of LLMs in recognizing ASCII art to bypass safety measures and elicit undesired behaviors from LLMs. ArtPrompt only requires black-box access to the victim LLMs, making it a practical attack. We evaluate ArtPrompt on five SOTA LLMs, and show that ArtPrompt can effectively and efficiently induce undesired behaviors from all five LLMs. Our code is available at https: //github.com/uw-nsl/ArtPrompt. 

In metals, orbital motions of conduction electrons on the Fermi surface are quantized in magnetic fields, which is manifested by quantum oscillations in electrical resistivity. This Landau quantization is generally absent in insulators. Here, we report a notable exception in an insulator—ytterbium dodecaboride (YbB 12 ). The resistivity of YbB 12 , which is of a much larger magnitude than the resistivity in metals, exhibits distinct quantum oscillations. These unconventional oscillations arise from the insulating bulk, even though the temperature dependence of the oscillation amplitude follows the conventional Fermi liquid theory of metals with a large effective mass. Quantum oscillations in the magnetic torque are also observed, albeit with a lighter effective mass. 

A<sc>bstract</sc> Thepp→W^±(→μ^±ν_μ)Xcross-sections are measured at a proton-proton centre-of-mass energy$$ \sqrt{s}=5.02 $$ $\sqrt{s}=5.02$TeV using a dataset corresponding to an integrated luminosity of 100 pb⁻¹recorded by the LHCb experiment. Considering muons in the pseudorapidity range 2.2< η <4.4, the cross-sections are measured differentially in twelve intervals of muon transverse momentum between 28< p_T<52 GeV. Integrated overp_T, the measured cross-sections are$$ {\displaystyle \begin{array}{c}{\sigma}_{W^{+}\to {\mu}^{+}{\nu}_{\mu }}=300.9\pm 2.4\pm 3.8\pm 6.0\ \textrm{pb},\\ {}{\sigma}_{W^{-}\to {\mu}^{-}{\overline{\nu}}_{\mu }}=236.9\pm 2.1\pm 2.7\pm 4.7\ \textrm{pb},\end{array}} $$ $\begin{array}{c}{\sigma }_{W^{+}\to {\mu }^{+}{\nu }_{\mu }}=300.9\pm 2.4\pm 3.8\pm 6.0\mathrm{pb},\\ {\sigma }_{W^{-}\to {\mu }^{-}{\overline{\nu }}_{\mu }}=236.9\pm 2.1\pm 2.7\pm 4.7\mathrm{pb},\end{array}$ where the first uncertainties are statistical, the second are systematic, and the third are associated with the luminosity calibration. These integrated results are consistent with theoretical predictions. This analysis introduces a new method to determine theW-boson mass using the measured differential cross-sections corrected for detector effects. The measurement is performed on this statistically limited dataset as a proof of principle and yields$$ {m}_W=80369\pm 130\pm 33\ \textrm{MeV}, $$ $m_W=80369\pm 130\pm 33\mathrm{MeV},$ where the first uncertainty is experimental and the second is theoretical. 

The first dedicated $Z$-boson mass measurement at the LHC with $Z\to {\mu }^{+}{\mu }^{-}$decays is reported. The dataset uses proton-proton collisions at a center-of-mass energy of 13 TeV, recorded in 2016 by the LHCb experiment, and corresponds to an integrated luminosity of $1.7{\mathrm{fb}}^{-1}$. A template fit to the ${\mu }^{+}{\mu }^{-}$mass distribution yields the following result for the $Z$-boson mass: $m_Z=91,185.7\pm 8.3\pm 3.9\mathrm{MeV}$, where the first uncertainty is statistical and the second systematic. This result is consistent with previous measurements and predictions from global electroweak fits.