China’s DePTH-GPT: A New AI Model to Revolutionize Deep-Sea Research

A team of Chinese scientists has launched a new artificial intelligence model named DePTH-GPT to advance deep-sea exploration. The model utilizes a combination of AI technologies, including deep learning, large language models, and computer vision, to analyze a wide range of data from the deep ocean such as video, topography, and bioacoustics. This new tool is expected to shift deep-sea research from traditional qualitative analysis to a more intelligent and predictive approach. DePTH-GPT has already been used to create an intelligent cognitive system for a seamount and a hydrothermal vent field. Developed as part of the UN Decade of Ocean Science for Sustainable Development, the model will be made available to global research institutions to create cognitive systems for various deep-sea habitats.

SETI Institute’s New AI Achieves 600x Speed Boost in Search for Alien Signals

Researchers at the SETI Institute, in collaboration with the Breakthrough Listen initiative and NVIDIA, have developed a new AI system that dramatically speeds up the search for Fast Radio Bursts (FRBs) from space. The system processes data 600 times faster than previous methods and operates 160 times faster than real-time. This significant increase in speed is achieved through an end-to-end AI architecture built on NVIDIA’s Holoscan platform, which analyzes streaming data in real-time without the need for traditional, time-consuming ‘dedispersion’ searches. The new system also improves accuracy by 7% and reduces false positives by a factor of ten. This technology could revolutionize the search for technosignatures, or potential signals from extraterrestrial civilizations, by enabling the detection of new and unexpected signal patterns.

Sources:

Google Research Tackles ‘Catastrophic Forgetting’ with New ‘Nested Learning’ Paradigm

Google Research has unveiled a new machine learning paradigm called ‘Nested Learning’ designed to address the issue of ‘catastrophic forgetting,’ where AI models lose proficiency on old tasks after learning new ones. The approach, detailed in a paper for NeurIPS 2025, views a single ML model as a system of interconnected, multi-level learning problems that are optimized simultaneously. This method treats the model’s architecture and its training algorithm as different levels of optimization, each with its own rate of update. By doing so, Nested Learning aims to create a more effective memory system for continual learning, drawing an analogy to the neuroplasticity of the human brain.

Sources:

Beep to Deploy Shared Autonomous Vehicle Fleets in Atlanta and Orlando by 2026

Beep Inc. has announced plans to launch two significant shared autonomous vehicle deployments in the United States in 2026. The services will be introduced in Atlanta, Georgia, and Altamonte Springs, Florida. In Atlanta, the autonomous vehicles will be deployed in collaboration with the Atlanta BeltLine Inc. to enhance mobility ahead of the 2026 FIFA World Cup. In Altamonte Springs, the existing CraneRIDES transit program will be expanded with autonomous vehicles to integrate with the regional rail network. The deployments will utilize the Karsan Autonomous e-JEST vehicle, ADASTEC’s Level-4 autonomous driving software, and Beep’s AutonomOS platform for fleet management and safety oversight. This initiative signals a move from pilot programs to scalable autonomous vehicle services integrated into public transit systems.

Sources:

Microsoft Enters AI Art Space with In-House Image Generator, MAI-Image-1

Microsoft has unveiled its first proprietary AI image generator, MAI-Image-1. The new tool is now accessible through Bing Image Creator and Copilot. According to the company, the model is particularly adept at rendering food, natural landscapes, and lighting effects. MAI-Image-1 is designed to balance the speed of image generation with the quality of the output, aiming to enhance creative productivity for its users.

Sources:

MIT CSAIL Develops New Framework for More Modular and Understandable Software

Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have introduced a new approach to software development designed to make code more modular, transparent, and easier to understand. The proposed framework addresses the issue of “feature fragmentation,” where the functionality of a single feature is scattered across multiple services within an application. The new model organizes software into “concepts,” which are self-contained pieces of code each designed for a specific task, and “synchronizations,” which are explicit rules defining how these concepts interact. This structure aims to make the relationships between different parts of the software visible and manageable, rather than hidden within low-level code. A domain-specific language (DSL) has been developed to express these synchronizations simply, in a format that large language models (LLMs) can reliably generate. The researchers believe this will lead to safer and more automated software development, as AI assistants could propose new features without introducing unforeseen side effects. In a case study, features like liking, commenting, and sharing were successfully centralized into individual concepts, making them easier to locate and test.

Sources:

‘formae’: A New Open-Source IaC Platform to Eliminate Cloud Drift

Platform Engineering Labs has announced the launch of ‘formae,’ a new open-source infrastructure-as-code (IaC) platform designed to address the inefficiencies of current IaC tools. The platform is built to be compiler-grade, type-safe, and drift-aware, automatically discovering and codifying an organization’s entire cloud environment into versioned, continuously synced code. Unlike traditional IaC tools that can become disconnected from the actual state of the infrastructure, formae serves as a live and accurate source of truth. It aims to reduce the manual effort and human error often associated with managing state files and complex workflows. The platform allows users to manage infrastructure changes as code, fitting into GitOps workflows and accommodating various tools for infrastructure modifications. By providing a unified view of all resources, formae is intended to help teams manage their cloud infrastructure with more confidence, mitigating issues like hidden drift and brittle states.

Sources:

Google Releases Magika 1.0, Rewritten in Rust for Faster AI File Detection

Google has officially released Magika 1.0, a significant update to its AI-powered file type detection system. The new version has been completely rewritten in Rust, resulting in substantial performance and security enhancements. Magika 1.0 now supports over 200 file types, a twofold increase from its initial release, with improved accuracy for text-based files like code and configuration. The system utilizes a compact deep-learning model to achieve high accuracy, which Google’s internal benchmarks place at around 99%. This release also includes a native Rust command-line client for improved speed and revamped Python and TypeScript modules for easier integration into other projects.

Sources:

OpenGuardrails Launches as an Open-Source Platform for LLM Safety

A new open-source project named OpenGuardrails has been introduced to enhance the safety of large language models in real-world applications. The platform provides a unified method for detecting unsafe, manipulated, or privacy-violating content. Developed by Thomas Wang and Haowen Li, OpenGuardrails allows organizations to define their own safety rules and adjust the model’s sensitivity to various risks. It uses a single large language model for both safety detection and manipulation defense, which streamlines deployment. The system is designed for enterprise use, capable of handling high traffic with low latency, and has demonstrated strong performance on English, Chinese, and multilingual benchmarks.

Sources:

The Hyvä Theme, a popular frontend for the Magento e-commerce platform, is being made open-source and available for free starting November 10, 2025. The theme will be relicensed under the Open Software License 3.0 (OSL-3.0) and the Academic Free License 3.0 (AFL-3.0), allowing developers to use, modify, and redistribute it without charge. This move is intended to lower the barrier to entry for merchants and developers, fostering greater community collaboration and innovation within the Magento ecosystem. While the core theme will be free, other products in the Hyvä suite, such as Hyvä Checkout and Hyvä Enterprise, will remain under a commercial license.

Sources: