Hardware
IonQ Targets Quantum Error-Correction Bottleneck With Single-CPU Decoder
IonQ has introduced a new quantum error decoder designed to reduce the classical computing overhead required for quantum error correction.
YESTERDAY, STILL WORTH KNOWING
Still
Crunchy.
A day of high-stakes hardware announcements, AI safety debates and a quiet breakthrough in biology.
Next-Gen Consoles and Gaming Peripherals
The race for the next generation of home gaming appears to have entered its final design phase. AMD leaker Kepler_L2 reports that both the PlayStation 6 and Microsoft's Xbox Helix have passed the 'taped out' stage, meaning their silicon is ready for mass production. The leak suggests the Xbox Helix will offer 56 teraflops of performance, while the PS6 is expected to hit 40 teraflops. This would give Microsoft a significant theoretical advantage over Sony's machine. Separately, Logitech has refined its flagship G Pro X3 Superstrike gaming mouse. The device retains its inductive clickers but swaps the microcontroller for a custom silicon chip, promising a 50 per cent improvement in battery life. The mouse also features a new weight distribution to address balance issues found in its predecessor.
AI Safety and Regulation
The conversation around artificial intelligence safety is moving from technical labs to high-level diplomacy. OpenAI CEO Sam Altman addressed the United Nations Security Council, urging global cooperation on AI safety and the maintenance of human control over advanced systems. This follows a report indicating that daily familiarity with AI does not necessarily reduce public concerns or support for regulation. The findings suggest that users may view the technology as inherently risky regardless of how often they interact with it. Meanwhile, the cybersecurity landscape is shifting as investors pour nine-figure sums into startups building defences for an AI-native world. This surge in funding is driven by fears of AI safety and the potential for rogue agents to cause harm.
AI in Action and Productivity
AI models are increasingly being deployed for practical, real-world tasks. NVIDIA has released a new model called Nemotron 3 designed to identify who is speaking in real-time conversations, a feature useful for meeting transcription and voice assistants. OpenAI has upgraded its ChatGPT Voice assistant to run on its new GPT-6 models and connect to email, calendar and Slack services, allowing users to handle tasks like booking meetings by voice. Ringg has launched AI agents powered by OpenAI's GPT-5.6 that resolve up to 65 per cent of customer calls across multiple channels. The company claims this system operates at 90 per cent of the cost of previous GPT-4.1 models. Google DeepMind has also introduced a persistent memory layer for its Private AI Compute platform, allowing AI assistants to maintain context across devices while keeping data encrypted and under user control.
Hardware Validation and Scientific Discovery
The reliability of AI hardware is being scrutinised by engineers at NVIDIA. Sakeena Fiza, a validation engineer, describes her work as detective work, hunting for the ways new AI systems might break before they reach customers. She details the process of bringing up new hardware, from initial power-up to rack-scale testing, ensuring components behave correctly under stress. This work highlights the interdisciplinary nature of modern hardware engineering. In a more surprising development, Anthropic's AI model Claude autonomously discovered a new enzyme system in a DNA database. The company claims this is the first result from its newly launched wet lab and an early test of Claude's usefulness for science. The discovery is similar to the machinery behind the gene-editing tool Crispr.