Artificial Intelligence & Machine Learning

Google’s AI Chief: ‘Learning to Learn’ is the Most Critical Future Skill

Demis Hassabis, CEO of Google’s DeepMind and a 2024 Nobel laureate, stated that the most crucial skill for the next generation will be ’learning how to learn’ to adapt to rapid advancements in Artificial Intelligence. Speaking in Athens, Hassabis highlighted the difficulty in predicting the future landscape given the weekly evolution of AI but asserted that significant change is inevitable. He suggested that artificial general intelligence (AGI) could be achieved within the next decade, leading to substantial progress. During the same event, Greek Prime Minister Kyriakos Mitsotakis expressed concerns that the immense wealth generated by a few large tech companies in the AI revolution could lead to significant social unrest if the benefits are not widely distributed.

Sources:

Autonomous Vehicles

Zoox Launches Public Robotaxi Service in Las Vegas

Amazon’s subsidiary, Zoox, has launched its public robotaxi service in Las Vegas, offering free rides in its purpose-built, fully autonomous vehicles. The service is accessible via the Zoox app and currently operates between a select number of destinations on and around the Las Vegas Strip, including Resorts World, Topgolf, and Area 15. The Zoox vehicle is uniquely designed without a steering wheel or pedals, featuring a carriage-style seating arrangement for four passengers. The company distinguishes itself by being the first to launch a ride-hailing service in a vehicle specifically built as a robotaxi. While the service is currently free, Zoox plans to introduce paid rides after receiving the necessary regulatory approvals. Following the Las Vegas launch, Zoox intends to expand its service to San Francisco, Austin, and Miami.

Sources:

Oxa Unveils Autonomous Electric Tow Tractor to Boost Logistics Efficiency

Autonomous vehicle technology company Oxa has introduced the Bradshaw T800, an autonomous electric tow tractor powered by its self-driving software. The T800 is designed for continuous operation in complex environments such as airports, warehouses, and manufacturing plants, with a capacity to tow up to eight tonnes. The vehicle aims to address labor shortages and increase efficiency in the logistics and manufacturing sectors. Key features include the ability to operate both indoors and outdoors, fire safety integration that moves the vehicle out of danger zones during an alarm, and dynamic zone management for controlling speed and restricted areas. A single operator can remotely monitor multiple vehicles, and the fleet can be switched between autonomous and manual driving modes.

Sources:

Robotics

Hengbot Showcases ‘Sirius’ AI Robotic Dog at IFA 2025

Global robotics company Hengbot presented its robotic dog, “Sirius,” at the IFA 2025 event in Berlin. Sirius is noted for its bionic design, which allows for lifelike movements, and its integration of emotional AI to foster a stronger connection with users. The robot features an AI head with up to 5 TOPS of edge computing power and utilizes large language models for interaction. It has a modular design and can be programmed using C, Python, or C++. Hengbot aims to create an intelligent robotics ecosystem with emotional connectivity, with Sirius being the first product in this vision. The company has already raised nearly $1 million for Sirius through a Kickstarter campaign.

Robotic Trio Successfully Tests Cooperative Lunar Cave Exploration

A team of European researchers has successfully demonstrated a cooperative robotic system for exploring lunar caves, which could serve as shelters for future astronauts. The test, conducted on the volcanic island of Lanzarote, Spain, involved a trio of robots with distinct roles. The mission concept, published in Science Robotics, unfolds in four phases: mapping the cave entrance, deploying a sensor cube for initial data collection, rappelling a scout rover into the cave, and finally, autonomous exploration and 3D mapping of the interior. The three robots involved were the SherpaTT, which acts as a surface anchor; the Coyote III, a scout rover that explores the cave; and the LUVMI-X, which assists with surface mapping and deploys the sensor cube. This robotic team approach is considered a safer and more cost-effective method for exploring these challenging extraterrestrial environments.

Sources:

Tools

Google Enhances Gemini App with Advanced Native Image Editing

Google has introduced a new native image generation and editing model within the Gemini app, developed by its DeepMind division. This update provides users with advanced capabilities to creatively alter images through simple text prompts. The tool allows for significant transformations, such as changing a person’s clothing to be made of different materials or redesigning a house into a tropical island theme. A key feature of the new model is its ability to keep subjects, like people or pets, consistent across multiple edits and to blend elements from different photos seamlessly. It can also modify specific parts of a picture while preserving the details in the rest of the image. These enhanced image editing features are now available for users to try in the Gemini app.

Sources:

DevSecOps

Cursor AI Code Editor Vulnerability Puts Developers at Risk

A significant security vulnerability has been discovered in the AI-powered code editor, Cursor. The flaw stems from the “Workspace Trust” feature being disabled by default, which can allow for the automatic execution of malicious code when a user opens a specially crafted repository. Security firm Oasis Security, which discovered the vulnerability, warns that this could lead to the leakage of sensitive credentials, file modification, or a wider system compromise, posing a serious risk of supply chain attacks. Unlike Visual Studio Code, on which Cursor is based, the disabled security setting means there is no prompt or user consent before code execution. To mitigate this vulnerability, users are advised to manually enable the Workspace Trust feature in Cursor, use a different editor for untrusted repositories, and audit new repositories before opening them.

Sources:

Quantum Computing

PsiQuantum Raises $1 Billion to Develop Million-Qubit Quantum Computers

PsiQuantum, a quantum computing startup, has successfully raised $1 billion in a Series E funding round. The new capital will be used to begin construction on quantum computing facilities in Brisbane, Australia, and Chicago, USA. This funding will also support the deployment of large-scale prototype systems and the advancement of their quantum photonic chips and fault-tolerant architecture. The round was led by affiliates of BlackRock, Temasek, and Baillie Gifford, with new investors including Nvidia’s NVentures unit. PsiQuantum is also collaborating with Nvidia on quantum algorithms, software, and the integration of GPUs with QPUs. The company aims to build fault-tolerant quantum computers with a million qubits.

Sources:

IonQ Gains UK Approval for Acquisition of Oxford Ionics

IonQ has received regulatory clearance from the UK’s Investment Security Unit for its acquisition of Oxford Ionics. The UK government has approved the acquisition under the condition that Oxford Ionics’ hardware and core operations remain in Britain. All of the company’s trapped-ion hardware must be hosted in the UK for independent assessment, and its staff, intellectual property, and manufacturing capabilities must also remain domestic. IonQ is a leading company in commercial quantum computing and networking, providing high-performance systems to solve complex problems. The company is working towards delivering quantum computers with 2 million qubits by 2030.

Sources:

Google’s Quantum Processor Used to Discover Exotic Phases of Matter

An international team of researchers has utilized Google’s Sycamore-class superconducting quantum processor to create and investigate a new phase of matter. This non-equilibrium phase, known as “Floquet topological order,” was realized in a system that was driven out of its natural equilibrium. The study, published in Nature, demonstrates how quantum computers can be used to explore highly entangled phases that are beyond the simulation capabilities of classical computers. The researchers observed key signatures of this exotic state, including chiral edge modes and anyonic excitations, in experiments that scaled up to 58 qubits. These findings highlight the potential of quantum processors as laboratories for new physics and could inform the development of more robust quantum error correction techniques.

Sources:

Software Engineering

MBZUAI and G42 Launch K2 Think, an Open-Source AI Reasoning System

The Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) and G42 have announced the release of K2 Think, a new open-source system for advanced AI reasoning. With 32 billion parameters, K2 Think is said to outperform larger reasoning models. This development is part of a series of UAE-developed open-source models and follows the 2024 release of the K2-65B foundation model. In a move towards greater transparency, K2 Think is fully open-source, including its training data, parameter weights, and software code for deployment and optimization. This allows the global research community to study, reproduce, and build upon the model’s reasoning processes. K2 Think will also be available on Cerebras’ wafer-scale compute platform to enhance reasoning performance and speed.

Sources: