Tootfinder

Since I fired up Chrome and enabled grid lanes to read Manuel’s post, I tried my 2019 reading order bookmarklet:
https://adrianroselli.com/2019/04/reading-order-bookmarklet.html
And it works!
You have to run it from the top of the page for the numbers t…

Leaky Language Models: Stealing Architecture and Inference Optimizations via Per-Token Timing
Sadegh Majidi, Niloofar Mireshghallah, Kazem Taram
https://arxiv.org/abs/2607.20723 https://arxiv.org/pdf/2607.20723 https://arxiv.org/html/2607.20723
arXiv:2607.20723v1 Announce Type: new
Abstract: This work presents LeakyLMs, a set of attacks that leak proprietary model, architecture, and deployment information from production language models. LeakyLMs is the first to demonstrate that key model and deployment details can be inferred using only token generation timing, even when interacting through remote APIs. LeakyLMs introduces two core attacks. The first attack targets inference optimizations and deployment strategies. For example, our attack detects whether a provider uses speculative decoding, a widely deployed inference-time optimization, and further identifies the context length of the draft model used in the pipeline. Our measurements show that Google Gemini Flash 2.5 uses speculative decoding with a draft context window of approximately 128K tokens. The second attack recovers key architectural properties, including the number of transformer layers, hidden dimension size, and number of attention heads. To achieve this, LeakyLMs builds a detailed and accurate model of token-generation timing on modern NVIDIA GPUs, characterizing how latency scales with model configuration and hardware parameters. The attack then performs a search over the architecture space using this timing model. In experiments with Llama models, the near-correct architectural configuration appears in the top-10 guesses more than 90% of the time.
toXiv_bot_toot

In case anyone else needs a helpful resource 😁
#photography #videography
https://markus-hagner-photography.com/

F Stop Chart (August 2026) Complete Guide to Aperture and Depth of Field - Markus Hagner Photography
Master f-stops with our comprehensive chart showing full, half, and third stops. Learn how aperture affects depth of field with practical examples for every photography type.

Reserved Top Level #DNS Names
https://datatracker.ietf.org/doc/html/rfc2606

Discretisation and independent resolutions of ample groupoids
Xin Li, Alistair Miller
https://arxiv.org/abs/2606.21761 https://arxiv.org/pdf/2606.21761 https://arxiv.org/html/2606.21761
arXiv:2606.21761v1 Announce Type: new
Abstract: We develop a general framework for understanding and computing both the groupoid homology of an ample groupoid and the topological K-theory of its reduced C*-algebra, based on two main ideas: discretisation and independent resolutions. Discretisation shows that a special class of ample groupoids we term independent groupoids are homologically and K-theoretically equivalent to discrete groupoids. We introduce the notion of a resolution by independent groupoids and provide a recipe for building a controlled independent resolution of a given ample groupoid of interest, leading to a systematic way of studying its homology and K-theory. In order to illustrate our general ideas and methods, we work out several concrete examples and applications. Garside categories provide a wide range of examples, including higher rank graphs, self-similar groups and spherical Artin-Tits groups. We also present an application to the homology of Stein's groups.
toXiv_bot_toot

OpenPINT: Open-source Planning for Isoeffective Nuclear Treatments in BNCT research
Ian Postuma, Sara J. Gonz\'alez, Setareh Fatemi, Cristina Pezzi, Carolina Ruzzon, Oreste Nicrosini, Valerio Vercesi, Silva Bortolussi
https://arxiv.org/abs/2606.21476 https://arxiv.org/pdf/2606.21476 https://arxiv.org/html/2606.21476
arXiv:2606.21476v1 Announce Type: new
Abstract: Objective: Present OpenPINT (Open-source Planning for Isoeffective Nuclear Treatments), an open-source treatment planning system for nuclear therapies that integrates Monte Carlo dose calculations with modular dosimetric and radiobiological models for photon-isoeffective dose evaluation.
Approach: We describe the software architecture, implementation choices, and data flow from segmented geometry and source configuration to NIfTI dose outputs. We define BNCT-relevant dosimetric metrics and evaluate the workflow with reproducible analytic and voxelized cylindrical-phantom benchmarks, supplemented by a geometric patient-positioning example.
Main results: The module provides a reproducible and scriptable path for generating MCNP-ready inputs, extracting component-wise BNCT dose maps, and computing analysis-ready outputs for quality checks and decision support. Fine-resolution voxelized configurations reproduced the 1 mm analytic reference within 0.13% for the brain-limited irradiation-time endpoint, whereas the full voxelized sweep exposed deviations up to 4.42% in coarse 8--10 mm configurations. Patient-wide gamma pass rates were at least 99.60% for the evaluated mesh/interpolation cases, while low-dose DVH-tail quantities remained sensitive to boundary discretization.
Significance: This first paper isolates and validates the simulation-preparation and dosimetric-analysis core of an open-source BNCT treatment-planning platform. It establishes a foundation for subsequent work on optimization, biological weighting, and clinical workflow integration.
toXiv_bot_toot

PhantomSeal: Proactive Deepfakes Defense with Identity/Context Protection and Forensic Tracing
Liangqin Ren, Zeyan Liu, Ye Wang, Yuxin Chen, Fengjun Li, Bo Luo
https://arxiv.org/abs/2607.20564 https://arxiv.org/pdf/2607.20564 https://arxiv.org/html/2607.20564
arXiv:2607.20564v1 Announce Type: new
Abstract: Deepfakes, especially face-swapping attacks, pose significant challenges to authenticity, security, and ethics across science, engineering, and society. While most existing detection/tracing approaches operate post hoc, proactive defenses that aim to intervene before deepfake generation remain limited in terms of real-world effectiveness. In this paper, we present PhantomSeal, the first proactive defense to simultaneously protect both the identity and the context of users' images from being used in face-swapping attacks, while supporting forensic tracing. We present a novel cloaking technique that embeds a selected identity as a stealthy identifier. This mechanism steers the deepfake generation process toward producing content that resembles the chosen cloak identity, thereby preventing successful face-swapping while enabling effective feature-based forensic analysis. The effectiveness and robustness of PhantomSeal is demonstrated in extensive experiments across different face-swapping architectures and models. For example, it reduces the attack success rate of SimSwap, an advanced deepfake model, to 0.30%, and correctly identifies 97.97% of manipulated content. Codes can be found at https://github.com/LiangqinRen/PhantomSeal
toXiv_bot_toot

Unterdessen bei der Digitalisierung: https://www.n-tv.de/panorama/Grossuebung-im-ICE-Tunnel-endet-im-grossen-Desaster-id31178818.html

"Längst alle tot", Fax-Kontakt: Großübung im ICE-Tunnel endet im großen Desaster
Bei einer Übung in einem Eisenbahntunnel in Thüringen trainieren Einsatzkräfte am Wochenende den Ernstfall. Doch bei einem simulierten ICE-Brand geht alles schief: Die Feuerwehr kommt zu spät, die Kommunikation läuft teilweise über Fax. Der Innenminister ist bedient.

Stell dir vor der ICE brennt, und die Feuerwehr kommt erst 2 Stunden später. Zum Glück nur eine Übung. Hoffen wir mal, dass man draus lernt…
https://www.n-tv.de/panorama/Grossuebung-im-ICE-Tunnel-endet-im-grossen-Desaster-id31178818.html

#Cache Stampede Prevention: Distributed Locking, Pub/Sub, and Request Coalescing
https://engineeringatscale.substack.com/p/cache-stampede-distributed-locking…