# Dec 13, 2025

## macOS 26.2 enables fast AI clusters with RDMA over Thunderbolt

- The **macOS Tahoe 26.2** release introduces significant enhancements, including improved **RDMA over Thunderbolt** support, which optimizes data transfer speeds and efficiency for developers. [Link to article](https://developer.apple.com/documentation/macos-release-notes/macos-26_2-release-notes#RDMA-over-Thunderbolt)

## We Built Another Object Storage (and Why It's Different)

- **FractalBits** addresses the **high-performance trap** in object storage by offering affordable, scalable solutions that support modern AI and analytics workloads, enabling nearly **1M GET/s** on small objects.

## Indexing 100M vectors in 20 minutes on PostgreSQL with 12GB RAM

- **VectorChord can now index 100 million 768-dimensional vectors in just 20 minutes on a 16 vCPU machine with only 12 GB of memory**, a significant improvement over pgvector, which requires 200 GB of memory and 40 hours for the same task.

## The Coming Need for Formal Specification

- **AI's role in coding is shifting** from implementation to generating tests and specifications, as models excel at creating unit tests from existing patterns in open-source code.

## Fast Median Filter over arbitrary datatypes

- The **median filter** is optimized through four versions, achieving a **420 times speedup** by utilizing an **ordinal transform** that efficiently manages pixel ranks instead of raw values, allowing for rapid median computation.

## [D] HTTP Anomaly Detection Research

- The project focuses on **anomaly detection of malicious HTTP requests** using a **NLP architecture** trained solely on benign samples, aiming to enhance firewall robustness against zero-day exploits.

## Improved Gemini audio models for powerful voice experiences

- The **Gemini 2.5 Flash Native Audio** upgrade enhances voice interactions by improving **function calling**, **instruction following**, and **conversation quality**, achieving a **71.5%** score on ComplexFuncBench for multi-step function calling.

## [R] [2512.01591] Scaling and context steer LLMs along the same computational path as the human brain

- **LLMs exhibit a striking alignment with human brain representations**, as evidenced by the study's analysis of brain signals during audiobook listening, revealing that initial LLM layers correspond to early brain responses while deeper layers align with later responses.

## [D] Parallel Reasoning Streams: Making LLMs Think Wider, Not Just Longer

- **Parallel reasoning streams** enhance LLMs by allowing them to explore multiple reasoning paths simultaneously, significantly improving efficiency and error correction compared to traditional sequential models.

## [D] GPT confidently generated a fake NeurIPS architecture. Loss function, code, the works. How does this get fixed?

## NeuralOperator Joins the PyTorch Ecosystem: Learning in Infinite Dimension with Neural Operators

- **NeuralOperator** joins the **PyTorch Ecosystem**, offering a dedicated library for learning **neural operators** that enables AI applications in **Science and Engineering**, particularly for solving **partial differential equations** through function space mappings.