Tootfinder

Attention Head Embeddings with Trainable Deep Kernels for Hallucination Detection in LLMs
Rodion Oblovatny, Alexandra Bazarova, Alexey Zaytsev
https://arxiv.org/abs/2506.09886

Attention Head Embeddings with Trainable Deep Kernels for Hallucination Detection in LLMs
We present a novel approach for detecting hallucinations in large language models (LLMs) by analyzing the probabilistic divergence between prompt and response hidden-state distributions. Counterintuitively, we find that hallucinated responses exhibit smaller deviations from their prompts compared to grounded responses, suggesting that hallucinations often arise from superficial rephrasing rather than substantive reasoning. Leveraging this insight, we propose a model-intrinsic detection method t…

I just saw the images of the DHS "rennovation" of the old St. Elizabeths in DC as the new DHS HQ.
The architectural renditions suggest that they will be turning it into something rather different than its historical shape (but the drawings i saw didn't look too bad - at least they didn't turn it into a Frank Gehry monstrosity with leaky roofs, dead-end hallways, and external stairs that require people to step open crevices.)
Given DHS head Noem's history, I …

"The destruction of the climate by industrial processes is a real-world problem. If programming language selection has no impact on it, then maybe we’re all just wasting our time, creating shiny consumer distractions to make the last few years of humanity that little more palatable for the richest 1% of the world’s human population."

Choosing A Programming Language
Selecting the correct programming language for a particular task often feels like a bunfight. Suggestions are thrown around based on the individual's preferences and desires. The experience is very likely to leave one feeling that programming is an art, rather than a science.

Traces of the evolution of cosmic void galaxies: An Integral Field Spectroscopy based analysis
Agust\'in M. Rodr\'iguez-Medrano, Dante J. Paz, Dami\'an Mast, Federico A. Stasyszyn, Andr\'es N. Ruiz
https://arxiv.org/abs/2506.07783

Traces of the evolution of cosmic void galaxies: An Integral Field Spectroscopy based analysis
Galaxies in the most underdense regions of the Universe, known as cosmic voids, exhibit astrophysical properties that suggest a distinct evolutionary path compared to galaxies in denser environments. Numerical simulations indicate that the assembly of void galaxies occurs later, leading to galaxies with younger stellar populations, low metallicities, and a high gas content in their halos, which provides the fuel to sustain elevated star formation activity. Our objective in this work is to test …

Series B, Episode 11 - Gambit
JENNA: She's right, Blake. He's led a privileged life. He's got money.
BLAKE: He's also been on the move for a long time.
JENNA: But even so, a man with his skills can always earn it.
https://blake.torpidity.net/m/211/170 B…

Blakes7Bot | Search
Companion site to the Blakes7Bot Twitter, Mastodon, and Bluesky accounts. Automated lines of Blakes 7 dialog. Includes a script search facility.

Series B, Episode 11 - Gambit
JENNA: She's right, Blake. He's led a privileged life. He's got money.
BLAKE: He's also been on the move for a long time.
JENNA: But even so, a man with his skills can always earn it.
https://blake.torpidity.net/m/211/170 B…

Blakes7Bot | Search
Companion site to the Blakes7Bot Twitter, Mastodon, and Bluesky accounts. Automated lines of Blakes 7 dialog. Includes a script search facility.

Stellar populations in STARFORGE II: Comparison with observations
Juan P. Farias, Stella S. R. Offner, Ronan Kerr, Michael Y. Grudi\'c
https://arxiv.org/abs/2506.00240

Stellar populations in STARFORGE II: Comparison with observations
Recent studies suggest that most star-forming regions in our Galaxy form stellar associations rather than bound clusters. We analyse models from the STARFORGE simulation suite, a set of magneto-hydrodynamical simulations that include all key stellar feedback and radiative processes following star formation through cloud dispersal. We create synthetic observations by introducing observational biases such as random spurious measurements, unresolved binaries, and photometric sensitivity. These bia…