#superintelligencelabs

BBC News | Meta workers can opt out of being tracked at work - but only for half an hour at a time AI generated summary, Read the full article for complete information. Meta is scaling back its plan to track employees’ computer activity after staff criticism of a tool that logged keystrokes and mouse clicks to train AI models. An internal memo revealed new controls will let workers pause data collection for up to 30 minutes at a time and request full exemptions, following a petition signed by more than 1,500 employees who called the initiative “dystopian.” The tool, part of Meta’s Model Capability Initiative, was originally justified as necessary for building AI agents that assist with everyday tasks, with assurances that the data would not be used for other purposes and that safeguards protected sensitive content. Employees, however, protested the privacy implications and reported increased battery drain and bandwidth usage, prompting Meta’s Superintelligence Labs VP Stephane Kasriel to announce optimizations and reaffirm confidence in the privacy protections while acknowledging concerns. This adjustment comes amid broader workforce reductions, including a planned 10 % cut of roughly 8,000 staff and 2,000 layoffs earlier this year. Read more: https://www.bbc.com/news/articles/c93x0k194yno?at_medium=RSS&at_campaign=rss #Meta #ModelCapabilityInitiative #StephaneKasriel #SuperintelligenceLabs

bing news | Mark Zuckerberg announces Muse Spark, a new Meta AI model: How to try it, benchmark results Mark Zuckerberg used a Wednesday Facebook post to unveil Muse Spark, the first model from Meta’s newly created Superintelligence Labs. Muse Spark powers the latest version of Meta AI, which is reachable at meta.ai or through the Meta AI app, and is billed as an “everyday personal‑use” assistant capable of visual understanding, health advice, shopping help, and social‑content creation. Zuckerberg framed the debut as the first step on Meta’s “scaling ladder” and hinted that future Muse models will act as autonomous agents that can perform tasks for users, with additional open‑source releases planned. The launch follows nine months of rapid development that began with a bold “personal superintelligence” vision laid out in a July 2025 manifesto. To build the lab, Meta recruited more than 50 researchers from rivals such as OpenAI, Anthropic, and Google, and tapped former Scale AI chief Alexandr Wang to lead the effort. After an aggressive hiring surge, the company paused hiring for budget planning and reorganized the team into four compact units focused on research, superintelligence development, products, and infrastructure—a structure Zuckerberg says fosters breakthrough work. Meta has earmarked $72 billion for AI in 2025 and expects to spend up to $135 billion in 2026, positioning Muse Spark as the first tangible product of these multi‑billion‑dollar investments. Meta released early benchmark scores for Muse Spark on tests such as Humanity’s Last Exam, ARC AGI 2, and GPQA Diamond, showing a mixed picture: the model outperforms some frontier systems (e.g., Claude Opus 4.6 Max, Gemini 3.1 Pro High, GPT 5.4 Xhigh, Grok 4.2) on certain tasks while lagging on others. An upcoming “Contemplating” mode, still in preview, will allow Muse Spark to orchestrate multiple reasoning agents in parallel, aiming to rival the extreme reasoning capabilities of competitors like Gemini Deep Think and GPT Pro. Users can try Muse Spark now via the web or the Meta AI mobile app, with a private API preview slated for select developers, and further feature rollouts will be announced gradually on meta.ai. Read more: https://mashable.com/article/mark-zuckerberg-meta-announce-new-muse-spark-ai-models #markzuckerberg #musespark #meta #superintelligencelabs #meta-ai