Nvidia shrinks LLM memory 20x without changing model weights

Nvidia's KV Cache Transform Coding (KVTC) compresses LLM key-value cache by 20x without model changes, cutting GPU memory costs and time-to-first-token by up to 8x for multi-turn AI applications.
Computer Weekly

Pathway builds truly native reasoning model to solve LLM Sudoku stumbling blocks

First set out in a scientific paper last September, Pathway’s post-transformer architecture, BDH (Dragon hatchling), gives LLMs native reasoning powers with intrinsic memory mechanisms that support ...
IEEE

RM-MOCO: A Fast-Solving Model for Neural Multi-Objective Combinatorial Optimization Based on Retention

Abstract: Multiobjective combinatorial optimization (MOCO) problems have a wide range of applications in the real world. Recently, learning-based methods have achieved good results in solving MOCO ...
Fox Business

GM recalls thousands of vehicles over transmission issue that could increase risk of crash

General Motors is recalling more than 43,000 Chevrolet, GMC and Cadillac SUVs due to a transmission issue that could potentially increase the risk of a crash. The recall affects 17,178 Chevrolet ...
IEEE

CNN-based Transistor Modeling under Self-Heating Effects

Abstract: Due to the steady shrinking of technology node sizes, Self-Heating Effect (SHE) has become at the forefront of reliability and longevity concerns. Without careful consideration of the ...
RealClearScience

The Top 5 Problems of the Standard Model

(via SciShow) The Standard Model of Particle Physics is the best theory scientists have to explain how the universe works on subatomic scales. But just because it's the "best" doesn't mean there ...
Live Science

'Proof by intimidation': AI is confidently solving 'impossible' math problems. But can it convince the world's top mathematicians?

AI could soon spew out hundreds of mathematical proofs that look "right" but contain hidden flaws, or proofs so complex we can't verify them. How will we know if they're right? When you purchase ...