Meta Muse briefly beat out ChatGPT on the iPhone free app store recently. I have been getting so many ads for Muse on my Instagram feed, and I just had to try it out. At the time of writing, it hasn’t yet completed my task, but I like what I’m seeing. I asked it to download my YouTube transcripts from the playlist, and it started downloading yt-dlp and got rate-limited by YouTube (this is good because it means it does the actual work). Meta Muse is not a traditional chatbot; it is supposed to connect to services like Gmail, Meta Analytics, and basically be an AI-powered version of IFTTT or Bixby Routines (AI smart automation). Meta also launched two physical products including a VR headset and a keychain like device for muse.
“After more than a decade building VR, we’re introducing Meta VR Glasses to define a new era for virtual reality. Launching in Spring 2027, this is our most advanced device yet, delivering a cinema, courtside seat, workspace, and console, all in a pair of glasses that weigh about 100 grams.”
At 1,299 dollars, it’s about half off Specs glasses, and they feature a much sleeker design, and you can watch movies in 3D.
“Meta VR Glasses are powered by Qualcomm’s new Snapdragon Reality Elite processor, enabling an exceptionally high-performance experience, fast load times, and breakthrough efficiency. The battery on the puck delivers up to three hours of continuous high-resolution media playback, and supports 45W fast charging.”
In Apple news (ugh), Apple released a new M5 Mac Studio that can run AI models locally. This is important to the AI debate, especially as lawmakers try to regulate AI. Will they regulate ultra-powerful chips such as the M5 series and RAM? I am not sure how that will hold up in a court of law, but I digress. Regardless, the Mac mini with the M5 Pro chip comes with 24 GB of unified memory (configurable up to 64 GB), 512 GB of storage (up to 8 terabytes), and Apple Intelligence. Of course, nobody is going to use Apple Intelligence on this when they have tools such as OpenClaw and open-source AI models at their disposal. An Ars Technica reporter downloaded LM Studio Bionic and ‘vibe coded’ (now a new word in the dictionary) to convert bank statements to spreadsheets. “First, I had the Qwen model write a Python script to convert bank statement PDFs into spreadsheets and format and sort them the way I wanted. I then built a web app around it so I could easily upload future sheets instead of messing with the LLM or the Python script.”
In Google News, Google has officially unveiled Project Suncatcher, an ambitious research moonshot aimed at exploring whether low Earth orbit (LEO) can host scalable machine learning infrastructure. With the inaugural prototype satellite scheduled to launch next week aboard SpaceX’s Transporter-18 rideshare mission in partnership with Planet, the initiative marks Google’s first major step toward testing AI hardware in space.
The Power Potential of Orbit
The primary catalyst behind Project Suncatcher is energy accessibility. Operating in low Earth orbit allows satellites to harness near-constant sunlight, generating up to eight times more solar power than equivalent systems on Earth. If scaled successfully, future satellite constellations could be linked together to manage massive AI workloads entirely in orbit, bypassing the severe energy and infrastructure constraints currently facing terrestrial data centers.
Solving the Engineering Hurdles
Moving AI compute beyond the atmosphere requires solving three formidable engineering challenges: hardware survival, thermal management in a vacuum, and ultra-high-bandwidth satellite interconnectivity.
- Hardware Survival: During a 10-minute rocket ascent, spacecraft experience sustained acceleration loads up to 10 times Earth’s gravity, with individual components enduring forces up to 50 to 100 g. Google subjected its Trillium Tensor Processing Units (TPUs) to rigorous vibration testing across all three axes to simulate launch conditions. Furthermore, to combat the harsh radiation of space, engineers tested the TPUs under active AI workloads at UC Davis’s Crocker Nuclear Laboratory proton beam facility, finding that the chips can survive a total ionizing dose exceeding a standard five-year space mission.
- Space Cooling: Operating in a vacuum means traditional convective cooling and airflow are non-existent; heat must be diffused exclusively via radiators and specialized heat pipes. The team has tested these thermal management systems extensively in thermal vacuum chambers and will use next week’s flight data to refine upcoming designs.
- Laser Interconnectivity: Future iterations of the satellite constellation will pack dozens of TPU chips per satellite, requiring constant high-bandwidth communication between units moving at high speeds. Because existing laser systems are generally optimized for low bandwidth over vast distances, Google is developing high-bandwidth, short-distance optical communication systems designed to maintain precision lock-ons while in motion—a capability slated for testing when two satellites launch in 2027.
Looking Ahead
Project Suncatcher represents a methodical, long-term roadmap for Google Research. By starting with survival validation and incremental orbital tests, the team aims to establish the foundational science required to ensure that future AI infrastructure can scale infinitely without straining Earth’s power grids. God bless and Tech Talk To You Later!!!
Leave a Reply