NerdNews

Apple will integrate OpenAI's GPT-5 model with Apple Intelligence, but it won't be available until iOS 26. Currently, Apple Intelligence uses GPT-4o. The update will bring improved AI capabilities to Apple devices.

OpenAI has unveiled GPT-5, retiring older models like GPT-4o and GPT-4.5. ChatGPT now features four distinct response styles, improved 'vibe-coding' skills, customization options, smarter voice mode, and easy Gmail & Google Calendar integration. GPT-5 also excels in creative writing and health-related topics, setting a new benchmark for AI assistance.

Former Google executive Mo Gawdat predicts a 15-year AI dystopia starting in 2027, citing AI's magnification of existing societal issues and 'human stupidity' as the primary drivers, rather than the technology itself. Gawdat believes humanity's values, such as freedom and human connection, will be disrupted, and that governments should regulate AI usage rather than the technology itself.

OpenAI has made two major announcements, releasing GPT-5, a powerful AI chatbot, and open-source models, in an effort to regain momentum in the AI industry after a summer of losing top talent and failing to acquire key startups. GPT-5 aims to tackle 'hallucinations' and provide more nuanced answers, while the open-source models are designed to appease developers and reassert OpenAI's relevance in the open AI ecosystem.

OpenAI's GPT-5 rollout was met with widespread criticism, prompting the company to bring back its older model, GPT-4o, for paid users. The move comes after just 24 hours of complaints, with many users expressing frustration over the new model's performance and the deletion of previous models.

Meta's Reality Labs Research previewed its latest VR prototypes, including the Tiramisu project with 'hyperrealistic VR' and Boba 3 headsets with ultrawide field of view. The Tiramisu headset has three times the contrast, 14 times the maximum brightness, and 3.6 times the angular resolution of the Meta Quest 3, but with a limited field of view and bulky form factor.

DJI has applied its drone obstacle avoidance system to a new line of robot vacuums called ROMO, featuring omnidirectional obstacle sensing, flexible cleaning arms, and high suction power. The vacuums also utilize machine learning algorithms for efficient cleaning routes and noise reduction systems. DJI is releasing the ROMO in three models, with prices starting at around $650, initially in China and later in overseas markets.

Brian Fargo discusses the use of AI in video game development, citing its potential for mundane tasks and inspiration, while expressing concerns about job losses and plagiarism. He suggests AI can be used ethically for tasks like data cleanup and voicing pickup lines, but not for generating entire games or art assets.

The winners of the AI Cybersecurity Challenge (AIxCC) were announced at DEFCON 33, with Team Atlanta taking the top prize of $4m. The challenge, funded by DARPA and ARPA-H, aimed to create AI-powered cybersecurity tools to secure US critical infrastructure and government services. The top three teams, including Trail of Bits and Theori, developed cyber reasoning systems that detected and patched vulnerabilities with high efficiency, demonstrating the potential of AI in cybersecurity.

Apple's goal for its Apple Intelligence-based Siri is to give it the ability to control apps through voice commands. This feature will allow users to perform tasks such as editing photos, buying items, or adding social media comments without touching the iPhone screen. However, it may exclude or limit the control of banking, financial, and other sensitive apps due to reliability concerns.

Microsoft will end support for Windows 10 on October 14, 2025. Users can upgrade to Windows 11 for free if their PC meets the minimum hardware specifications, buy a new PC, continue using Windows 10 without security updates, or pay for extended security updates until October 13, 2026.

The Uncut Gem project introduces a fully open-source quantum sensor that uses a special diamond to detect magnetic and electrical fields, making quantum technology more accessible and affordable for researchers and enthusiasts.

The Australian Information Commissioner has launched civil action against Optus for a 2022 data breach that exposed 9.5 million Australians' personal details. The lawsuit alleges Optus failed to protect victims' personal information, breaching Australia's Privacy Act 1988.

The DEF CON Franklin project has been working to improve the cybersecurity of US water systems. After a successful pilot program, the project is now expanding to protect thousands of water utilities across the country. The project provides free cybersecurity services, including asset inventories, operational technology assessments, and network mapping and scanning. The goal is to protect these critical infrastructure systems from cyber threats, particularly from Chinese and Iranian hackers.

A sudden spike in demand for compute resources in Azure's East US region caused allocation failures for virtual machines, despite Microsoft claiming the issue is resolved. The problem, which began on July 29, was caused by insufficient capacity and affected several generations of Intel and AMD VMs.

Security researchers discovered two techniques to crack high-security electronic safes with Securam ProLogic locks, used by multiple brands. The locks can be opened in seconds, and the vulnerabilities affect at least eight brands of safes. The researchers found that the locks have a reset method and a hidden debug port that can be exploited to gain access to the safe's unlock code.

A Chinese company, GoLaxy, has been uncovered using AI to generate social media profiles and influence public opinion in the US. The company has been targeting US politicians and social media influencers, and has been linked to the Chinese government. Former NSA boss General Paul Nakasone has warned that China is upping its propaganda game and that the use of AI is a new and evolving threat.

A former White House cyber and counter-terrorism expert has criticized Microsoft for not taking security seriously, citing the company's numerous security vulnerabilities and exploits, including recent incidents involving SharePoint and Exchange Server, which have been targeted by Chinese and Russian spies.

Researchers found a prompt injection vulnerability in Google's Gemini large language model-powered applications, allowing for attacks such as smart home system takeover and email exfiltration. The issue has been fixed by Google after responsible disclosure.

A Kentucky town used AI to gather opinions from nearly 8,000 residents, finding that 80% agreed on 2,370 ideas, including increasing healthcare specialists and repurposing empty retail spaces. The AI tool Sensemaker analyzed the responses, saving the county an average of 28 work days.