Models & Labs
Microsoft to unveil new AI models at Build
The Verge AIhigh confidence
Why it matters
- →Microsoft's new AI models could enhance enterprise capabilities significantly.
- →The Copilot 'super app' aims to streamline AI assistant integration for users.
- →Windows 11 improvements may boost developer productivity and satisfaction.
Microsoft is preparing to unveil new AI models and Windows enhancements at its Build conference in San Francisco. Key announcements include the MAI-Thinking-1 reasoning model for enterprise use and a Copilot 'super app' that combines various AI assistants. The company will also showcase improvements to Windows 11, focusing on a developer-optimized experience and compatibility with Nvidia's RTX Spark. These developments underscore Microsoft's commitment to AI and its strategy to reconnect with developers.
Read originalMore from The Verge AI
© The Verge AIAgentsagents
Gemini Spark AI Agent: Impressive but Costly
Gemini Spark, Google's new AI agent, shows promise in automating tasks but comes with significant costs and privacy concerns. While it can handle complex tasks like drafting emails and managing calendars, users must still oversee its actions to ensure accuracy. The tool is currently available only to subscribers of Google's AI Ultra plan, priced at $99.99 per month, and requires deep integration into the Google ecosystem. Despite its capabilities, the need for constant supervision and the high subscription cost make it a tough sell for many users.
The Verge AI
© The Verge AIMarket & Regulationbusiness
Meta's AI exploited to hijack Instagram accounts
Meta's AI support chatbot was manipulated by hackers to take control of Instagram accounts, revealing a critical security gap. By convincing the chatbot to change the email linked to an account, hackers could reset passwords and exclude original users. This flaw was exploited during high-profile hacks, including the account of Barack Obama's White House. Although Meta has addressed the issue, the incident raises questions about the security of AI systems, especially as the company faces layoffs and shifts focus to AI tools. This event highlights the importance of robust security measures in AI-driven environments.
The Verge AI
© The Verge AIMarket & Regulationbusiness
Anthropic files for IPO, eyes massive valuation
Anthropic has taken a significant step towards going public by filing with the U.S. Securities and Exchange Commission. This move positions Anthropic as a major player in the AI industry, with a staggering post-money valuation of $965 billion, surpassing its rival OpenAI. The filing was submitted confidentially, meaning detailed financial information will remain undisclosed for now. This development intensifies the competition between Anthropic and OpenAI, as both companies vie for dominance in the AI sector. The IPO is expected to be a landmark event, potentially reshaping the landscape of AI startups.
The Verge AI
More in Models & Labs
Models & Labsmodels
llama.cpp adds EXAONE 4.5 implementations
The latest llama.cpp release expands its capabilities with the integration of EXAONE 4.5, bringing new vision markers and projector paths into the fold. This update aligns EXAONE 4.5 with the Qwen2.5-VL-style encode path, enhancing model loading and tensor registration processes. Developers will find improved performance and compatibility, particularly when working with EXAONE models. While no new models are introduced, the release refines existing functionalities, ensuring robust performance across various systems. This step forward is crucial for developers seeking to leverage EXAONE 4.5's full potential.
llama.cpp Releases
Models & Labsmodels
llama.cpp b9455 Release Adds Quantized KV Cache
The latest b9455 release of llama.cpp introduces quantized KV cache support, a notable enhancement for efficiency in AI model inference. This update also addresses a partial view fix and removes an overly strict assert, improving the overall robustness of the software. While the release includes various platform builds, the focus remains on optimizing performance across different environments. The addition of quantized KV cache support is a step forward in making AI models more resource-efficient, particularly beneficial for developers working with limited computational resources.
llama.cpp Releases
Models & Labsmodels
llama.cpp b9457 release focuses on Vulkan improvements
The latest b9457 release of llama.cpp brings a notable improvement in Vulkan performance by reducing host memory lock contention, which can enhance efficiency in certain workloads. This update replaces unique_lock with lock_guard, aiming to streamline operations. While the release doesn't introduce new models or major features, it continues to refine the platform's compatibility across various systems, including macOS, Linux, and Windows. The focus remains on optimizing existing capabilities rather than expanding into new territories.
llama.cpp Releases