MIT Tech Review
Chinese tech workers are starting to train their AI doubles–and pushing back
Chinese tech workers are being ordered by management to build AI versions of themselves — essentially training their own replacements. The directive is sparking rare pushback from a workforce that has largely embraced AI tools, forcing workers to confront the uncomfortable reality of automation on a personal level. The tension signals a broader reckoning in the industry as enthusiasm for AI collides with job security fears.
Read article →
The Verge
Cloud development platform Vercel was hacked
Vercel, a widely used platform for hosting and deploying web applications, suffered a security breach linked to the ShinyHunters hacking group, which also targeted Rockstar Games. Stolen data including employee names, email addresses, and activity timestamps has been posted online, with hackers reportedly attempting to sell additional information. Vercel confirmed the incident, stating it affected a limited subset of customers and pointing to a compromised third-party vendor as the entry point.
Read article →Hacker News
Uber’s Anthropic AI push hits a wall
Uber's attempt to integrate Anthropic's AI into its operations has run into significant obstacles, raising questions about the practical limits of deploying large language models in complex, real-world logistics environments. The stumble highlights a broader challenge facing enterprises rushing to embed cutting-edge AI into mission-critical workflows — the gap between demo-room promise and production-grade reliability. For Uber, which processes millions of rides and deliveries daily, the stakes of getting AI integration wrong are anything but theoretical.
Read article →Hacker News
Show HN: TRELLIS.2 image-to-3D running on Mac Silicon – no Nvidia GPU needed
Microsoft's 4-billion-parameter TRELLIS.2 image-to-3D model has been ported to run natively on Apple Silicon, eliminating the hard dependency on Nvidia CUDA infrastructure. The developer replaced core CUDA-specific operations — including sparse convolutions, flash attention, and GPU-accelerated mesh extraction — with pure-PyTorch equivalents across roughly a few hundred lines of changes. The port opens high-quality 3D asset generation to the tens of millions of Mac users who previously had no viable path to running the model locally.
Read article →Hacker News
Show HN: Run TRELLIS.2 Image-to-3D generation natively on Apple Silicon
Microsoft's TRELLIS.2, a 4-billion parameter image-to-3D model previously locked behind CUDA-dependent libraries, can now run natively on Apple Silicon thanks to a community port using PyTorch MPS. The developer replaced CUDA-specific operations — including flash attention, custom sparse convolution kernels, and GPU-based mesh extraction — with pure-PyTorch alternatives across fewer than a thousand lines of changes. The port opens serious 3D generation capabilities to Mac users without requiring a separate Linux or Windows machine with an Nvidia GPU.
Read article →Get this delivered every morning
Join thousands of readers who get the world's most important stories, curated daily.
Start reading free →