PrismML has launched Bonsai 27B, a 27-billion-parameter model based on Alibaba’s Qwen 3.6 that utilizes extreme quantization to run locally on consumer hardware. The release includes a 3.9 GB 1-bit binary variant designed for mobile devices and a 5.9 GB ternary variant optimized for laptops. By reducing weights to just one or two bits, these models bypass the high memory and bandwidth requirements that typically restrict large-scale AI to cloud infrastructure.

Source

Apple is exploring a partnership with PrismML, a Caltech spinout, to integrate extreme model-compression technology into future iPhone hardware. The startup utilizes ultra-low-bit weight architectures, such as 1-bit and ternary representations, to drastically reduce the memory footprint of large language models. This mathematical approach allows a 27-billion-parameter model like Alibaba’s Qwen 3.6 to shrink from 54 GB to under 4 GB, making it compatible with the 8 GB of RAM found in current Pro-tier iPhones.

Source

Apple has released the iOS 27 public beta, introducing a rebuilt Siri AI that functions as a dedicated chatbot app powered by Apple Intelligence. This update enables the assistant to leverage a semantic index of local data, including emails, messages, and photos, to provide personalized context for user queries. Unlike third-party chatbots, the new Siri can execute complex in-app actions and process multiple simultaneous commands while maintaining a straightforward, task-oriented tone.

Source

Mozilla has announced plans to shift Firefox for desktop and Android to a bi-weekly release cadence beginning in September 2026. This experimental move transitions the browser from its current four-week cycle to a faster schedule to deliver ready-to-ship features more frequently. The change officially begins with the release of Firefox 155 on September 1, 2026, following the final monthly update of version 154 in August.

Source

Windows 11 users are reporting a significant memory leak tied to the Cross Device Service, a background component used by the Phone Link application. This service facilitates features like notification syncing and clipboard sharing but can erroneously consume between 15GB and 30GB of RAM. The issue appears to develop gradually over time, eventually degrading system performance and forcing users to manually terminate the process via Task Manager.

Source

Deutsche Telekom is undergoing a fundamental transformation to become an AI-native telecommunications provider by embedding generative AI into its core infrastructure. The company has moved beyond simple productivity tools, instead redesigning its entire operating model to automate decision-making and service delivery. This strategy involves a massive scale-up of AI tool usage among employees and the integration of advanced models into daily customer interactions.

Source

Instagram head Adam Mosseri recently stated that the platform will not filter out AI-generated content but will instead focus on labeling it for transparency. He suggested that users who dislike AI content should have the ability to remove it from their feeds through personalized settings. This approach aligns with Meta's broader strategy to move toward highly individualized, algorithmically driven experiences for every user.

Source