OpenAI Unveils Voice Assistant

Today:

• OpenAI Unveils Voice Assistant

• Google I/O 2024

• Apple Enhances Siri for iOS 18

• Fugaku Supercomputer Powers New Japanese AI

• AI Transcription Coming to Apple Voice Memos

• Arm Holdings to Debut AI Chips in 2025

•  'LLMjacking' Attacks Cloud AI Services

OpenAI Develops AI Voice Assistant As It Chases Google, Apple

OpenAI is advancing its AI capabilities to create a voice assistant that surpasses today's technology. This development includes enabling the AI to communicate through sound and text and recognize objects and images. The enhancements, witnessed by select customers, aim for more intuitive and logical interactions than existing models. 

OpenAI's CEO, Sam Altman, envisions a future where AI can seamlessly integrate into daily life, similar to the film "Her." A public demonstration might occur on Monday, potentially positioning OpenAI ahead of Google, which plans its own AI announcements later that week.

THE INFORMATION

Google I/O 2024 will be all about AI again

Google I/O 2024, starting on May 14th, is set to focus extensively on artificial intelligence (AI). Since last year's conference, Google has introduced Gemini, a powerful AI model, and has been enhancing its applications in Google Search, Maps, and Android. This year’s conference will explore how Google integrates AI with these services, including new features like AI-driven conversation tools for English learners, image generation for shopping, and AI support in finding electric vehicle chargers on Google Maps. 

Another highlight might be the debut of "Pixie," a new digital assistant powered by Gemini, expected to add innovative multimodal interactions to Google’s device ecosystem. Hardware announcements may be limited, given recent product launches, but leaks suggest potential teasers for the Pixel 9 and a rebranded Pixel Fold.

THE VERGE

Siri in iOS 18 rumored to be 'more conversational and versatile' thanks to AI

Siri is anticipated to become more conversational and versatile in iOS 18. As Apple prepares for its WWDC 2024, it has refocused efforts from other projects like the Apple Car to enhance Siri. The upgrade aims to make Siri more dynamic in conversations, rather than just responding to isolated commands. 

While it won’t directly compete with services like ChatGPT, Siri will improve in its existing functions such as setting timers and scheduling. Additionally, Apple is reportedly developing its own large language models for local and cloud processing, and may integrate a third-party chatbot like ChatGPT or Google Gemini into its systems.

TECHRADAR

Japan team uses Fugaku supercomputer to develop language model for AI

A team from Tokyo Institute of Technology and various collaborators, including Fujitsu, developed a language model using Japan's supercomputer Fugaku. This model, primarily trained on Japanese data, aims to advance AI applications tailored for local use. 

Unlike typical models which rely on continual updates, Fugaku-LLM was built from scratch for clearer learning processes and safer outcomes. It's also unique because it operates on CPUs rather than the commonly used GPUs, showcasing Fugaku's enhanced capabilities. This development highlights a significant move towards self-reliance in technology, avoiding dependence on foreign AI systems. The model's source code is publicly available on Fujitsu's website.

THE JAPAN TIMES

Apple might bring AI transcription to Voice Memos and Notes

Apple is planning to introduce AI transcription capabilities in its Voice Memos and Notes apps with iOS 18. This feature will allow real-time transcription of audio recordings, potentially saving users from having to listen to them again. While the Notes app currently doesn't support audio recording, this is also expected to change with the upcoming update. 

Additionally, Apple may enhance the Notes app by adding AI-powered summaries of recorded audio. These upgrades aim to benefit students, journalists, and anyone who frequently uses audio notes. Apple emphasizes privacy in its AI features, processing simpler requests on the device and outsourcing more complex tasks to protect user data. 

THE VERGE

Arm Holdings plans to launch AI chips in 2025

Arm Holdings, owned by SoftBank Group, is set to enter the AI chip market by 2025. The UK-based company plans to establish an AI chip division, aiming to create a prototype by spring 2025 and start mass production by autumn the same year. The development costs, potentially in the hundreds of billions of yen, will be initially funded by Arm, with financial support from SoftBank. 

Discussions are underway with Taiwan Semiconductor Manufacturing Corp (TSMC) to secure manufacturing capacity. Arm's move into AI chips is part of a broader strategy to expand into the data-center market and reduce dependency on Nvidia, a dominant player in the industry.

REUTERS

Researchers Uncover 'LLMjacking' Scheme Targeting Cloud-Hosted AI Models

Cybersecurity experts have identified a new cyberattack called "LLMjacking," which uses stolen cloud credentials to infiltrate cloud-hosted AI services like large language models (LLMs). The Sysdig Threat Research Team reported that attackers first breach systems using vulnerabilities, like those in the Laravel Framework, to acquire cloud service credentials. They then manipulate these credentials to access and control LLM services without detection, essentially selling unauthorized access to these resources. 

The attack could result in substantial costs for the victims, potentially exceeding $46,000 daily. Experts advise organizations to enhance logging and monitoring of cloud activities and strengthen vulnerability management to prevent such breaches.

THE HACKER NEWS