Tootfinder

Learning Obfuscations Of LLM Embedding Sequences: Stained Glass Transform
Jay Roberts, Kyle Mylonakis, Sidhartha Roy, Kaan Kale
https://arxiv.org/abs/2506.09452

Learning Obfuscations Of LLM Embedding Sequences: Stained Glass Transform
The high cost of ownership of AI compute infrastructure and challenges of robust serving of large language models (LLMs) has led to a surge in managed Model-as-a-service deployments. Even when enterprises choose on-premises deployments, the compute infrastructure is typically shared across many teams in order to maximize the return on investment. In both scenarios the deployed models operate only on plaintext data, and so enterprise data owners must allow their data to appear in plaintext on a …

The relative index in coarse index theory and submanifold obstructions to uniform positive scalar curvature
Alexander Engel, Christopher Wulff
https://arxiv.org/abs/2506.14301

The relative index in coarse index theory and submanifold obstructions to uniform positive scalar curvature
We provide a coarse version of the relative index of Gromov and Lawson and thoroughly establish all of its basic properties. As an application, we discuss a general procedure to construct wrong way maps on the $K$-theory of the Roe algebra mapping the coarse index class of the Dirac operator of a manifold to the one of a suitably embedded submanifold of arbitrary codimension, thereby establishing an abstract machinery to find obstructions to uniform positive scalar curvature coming from these s…

The Trump administration’s deployment of national guard troops to Los Angeles
to intervene in civilian protests in the face of opposition from the Californian governor
is a major escalation that risks the politicisation of the US military, armed service veterans are warning.
Former top military figures have told the Guardian that the decision to put up to 2,000 troops under federal control and send them into the streets of LA
is a violation of the military’s commitmen…

A triple torsion linking form and 3-manifolds in $S^4$
Michael Freedman, Vyacheslav Krushkal
https://arxiv.org/abs/2506.11941 https://

A triple torsion linking form and 3-manifolds in $S^4$
Given a rational homology 3-sphere $M$, we introduce a triple linking form on $H_1(M; \mathbb{Z})$, defined when the classical torsion linking pairing of three homology classes vanishes pairwise. If $M$ is the boundary of a simply-connected 4-manifold $N$, the triple linking form can be computed in terms of the higher order intersection form on $N$, introduced by Matsumoto. We use these methods to formulate an embedding obstruction for rational homology spheres in $S^4$, extending a 1938 theore…

Factorization method for near-field inverse scattering problems in elastodynamics
Chun Liu, Guanghui Hu, Tao Yin, Bo Zhang
https://arxiv.org/abs/2505.24288

Factorization method for near-field inverse scattering problems in elastodynamics
Consider a time-harmonic elastic point source incident on a bounded obstacle which is embedded in an open space filled with a homogeneous and isotropic elastic medium. This paper is concerned with the inverse problem of recovering the location and shape of the obstacle from near-field data generated by infinitely many incident point source waves at a fixed energy. The incident point sources and the receivers for recording scattered signals are both located on a spherical closed surface, on whic…

This https://arxiv.org/abs/2411.06426 has been replaced.
initial toot: https://mastoxiv.page/@arXiv_csCR_…

SequentialBreak: Large Language Models Can be Fooled by Embedding Jailbreak Prompts into Sequential Prompt Chains
As the integration of the Large Language Models (LLMs) into various applications increases, so does their susceptibility to misuse, raising significant security concerns. Numerous jailbreak attacks have been proposed to assess the security defense of LLMs. Current jailbreak attacks mainly rely on scenario camouflage, prompt obfuscation, prompt optimization, and prompt iterative optimization to conceal malicious prompts. In particular, sequential prompt chains in a single query can lead LLMs to …

This https://arxiv.org/abs/2503.05572 has been replaced.
initial toot: https://mastoxiv.page/@arXiv_mat…

Word problems and embedding-obstructions in cellular automata groups on groups
We study groups of reversible cellular automata, or CA groups, on groups. More generally, we consider automorphism groups of subshifts of finite type on groups. It is known that word problems of CA groups on virtually nilpotent groups are in co-NP, and can be co-NP-hard. We show that under the Gap Conjecture of Grigorchuk, their word problems are PSPACE-hard on all other groups. On free and surface groups, we show that they are indeed always in PSPACE. On a group with co-NEXPTIME word problem, …