AI News

Follow the latest AI and automation news

10 Aug 202627 Safar 1448 AH

Meta returns to open models with Zuckerberg's plan to out-copy China and sell compute by auction

Meta has announced the launch of Muse Glimmer, its first open model from Superintelligence Labs, featuring a size of 30 billion parameters. The model can run on consumer hardware after weight compression, requiring less than 20 GB of memory. In an accompanying essay, Mark Zuckerberg defends the practice of distilling other labs' models and calls for fewer restrictions on US labs. This development reflects Meta's strategy to compete with major players like OpenAI and Anthropic, with an open-weight version of Muse Spark 1.2 expected to follow soon. This could reshape the AI landscape and enhance Meta's ability to auction compute resources.

Infrastructure

10 Aug 202627 Safar 1448 AH

Microsoft to Unveil Maya 300 AI Chip This Fall

Microsoft is set to officially unveil its new AI chip, the Maya 300, this fall, potentially as soon as next month, according to The Information. The company initially revealed the Maya chip in November 2023 but has lagged behind competitors in scaling production, aiming to reduce its reliance on costly Nvidia processors. Microsoft is in talks with TSMC to secure production capacity for over 300,000 units of the chip, expected to be delivered by 2027, highlighting Microsoft's ambitions to strengthen its position in the cloud computing market.

Business

10 Aug 202627 Safar 1448 AH

Intense Competition in AI Market Between US and Chinese Companies

US tech giants in the AI sector are facing fierce competition from Chinese counterparts, with new low-cost models emerging. OpenAI has reduced the usage fees for its lightweight models (Luna) by 80%, while Anthropic introduced a competitively priced new model. Conversely, some Chinese firms are increasing their prices amid rising demand for their open-source models.

💡

'ruya Bank' goes live with Agentic AI services

How might AI reshape the future of Islamic banking?

Models

Model ML completes finance work more efficiently with GPT-5.6 Sol

10 Aug 202627 Safar 1448 AH

Models

10 Aug 202627 Safar 1448 AH

Model ML completes finance work more efficiently with GPT-5.6 Sol

The GPT-5.6 Sol model has been launched to enhance financial operations, enabling advanced financial research and analysis. This model allows for the creation of editable PowerPoint presentations and traceable Excel workbooks, facilitating decision-making processes. This development represents a significant step in improving financial efficiency, saving time and resources for executives in making data-driven strategic decisions.

Business

Nvidia and Zayo collaborate on network capacity

10 Aug 202627 Safar 1448 AH

Business

10 Aug 202627 Safar 1448 AH

Nvidia and Zayo collaborate on network capacity

Nvidia has announced its partnership with Zayo to build critical network capacity for AI factories. Zayo will construct over 8,000 miles of long-haul fiber across fast-growing AI corridors, significantly expanding its existing infrastructure. The initiative spans 24 routes, including six new long-haul routes and 10 overbuilds of existing networks, reflecting a growing response to compute demand. Zayo's CEO, Steve Smith, noted that AI is reshaping how network infrastructure is built across the U.S.

Governance

The AI Slop Backlash Is Actually Having an Impact

10 Aug 202627 Safar 1448 AH

Governance

10 Aug 202627 Safar 1448 AH

The AI Slop Backlash Is Actually Having an Impact

An increasing number of platforms are recognizing that users do not want to consume low-quality AI-generated content. Many sites and apps are implementing tools and policies aimed at flagging, labeling, and banning AI-generated material. These measures indicate a significant shift in how platforms handle content, aiming to provide a better user experience by ensuring content quality. These policies include effective mechanisms for identifying AI-generated content, reflecting a growing awareness of the importance of verifying information sources.

Models

ByteDance is looking to develop an AI model to rival Mythos

10 Aug 202627 Safar 1448 AH

Models

10 Aug 202627 Safar 1448 AH

ByteDance is looking to develop an AI model to rival Mythos

Recent reports indicate that ByteDance is developing an AI model that could rival Anthropic's Mythos system. The company is in the early stages of training a model with up to 10 trillion parameters, making it three times larger than Moonshot Kimi K3, the largest model from China to date. ByteDance aims to strengthen its position in China's AI ecosystem with this model. According to the Financial Times, Anthropic's Mythos 5 has 8 trillion parameters, while its Fable 5 has about 5 trillion. ByteDance's model is reportedly in the pre-training phase, which typically takes three to six months.

Business

SpaceX to develop a new semiconductor facility in the U.S.

10 Aug 202627 Safar 1448 AH

Business

10 Aug 202627 Safar 1448 AH

SpaceX to develop a new semiconductor facility in the U.S.

SpaceX has announced plans to build a massive semiconductor fabrication facility in Texas, with an initial investment exceeding $16.8 billion. The Terafab project is expected to create 3,000 jobs and is backed by a $30 million grant from the Texas Enterprise Fund. The 100-million-square-foot site in Grimes County will combine logic, memory, and advanced packaging, allowing SpaceX to consolidate parts of the semiconductor production chain. This facility will enable the company to produce AI chips at scale for use both on Earth and in space.

Agents

Meta Muse Glimmer brings local AI agents to consumer GPUs

10 Aug 202627 Safar 1448 AH

Agents

10 Aug 202627 Safar 1448 AH

Meta Muse Glimmer brings local AI agents to consumer GPUs

Meta has announced the release of Muse Glimmer under an Apache 2.0 license, designed for local AI agents that can run on consumer GPUs. The 30-billion-parameter model allows developers to utilize it for local coding and function calling, addressing the operational constraints of cloud-hosted models that require network access. Meta's benchmark tests indicate that Muse Glimmer outperforms other models like Gemma4-31B and Qwen3.6-27B across five out of eight general-agent benchmarks. The model scored 75.5 on the MCP Atlas test, showcasing its capability to handle multi-turn requests effectively.