PLUS: Meta cuts metaverse spending for AI, and inside Anthropic's AI-powered workplace
Good morning
An open-source AI is now generating GPU code that runs faster than NVIDIA’s own specialized, hand-optimized software. The project uses reinforcement learning to find the most efficient ways to structure core AI operations.
This development suggests a future where developers can automatically extract maximum performance from complex hardware. It raises a key question: how will the role of expert programmers change when AI can consistently optimize code better than humans?
In today’s Next in AI:
An open-source AI beats NVIDIA's own code
Meta pivots from metaverse to AI hardware
Inside Anthropic's AI-powered workplace
NVIDIA CEO on AI augmenting jobs
AI Beats NVIDIA's Code with... More AI
Next in AI: An open-source AI system is writing GPU code that runs faster than NVIDIA's own hand-optimized libraries for matrix multiplication, a core task for AI workloads. The new research paper details how the system, called CUDA-L2, uses reinforcement learning to achieve this.
Decoded:
CUDA-L2 combines large language models with reinforcement learning to automatically discover the most efficient ways to structure complex GPU operations.
In testing, the AI-generated code consistently outperforms NVIDIA's proprietary
cuBLASlibrary on A100 GPUs across 1,000 different configurations.The team plans to extend support to newer GPUs like Hopper and Blackwell, suggesting AI-driven code optimization could become a standard for future hardware.
Why It Matters: This project shows that AI can automate highly specialized, performance-critical programming tasks traditionally reserved for expert engineers. It signals a future where developers can rely on AI to extract maximum performance from complex hardware automatically.
Meta Pivots from Metaverse to AI
Next in AI: Meta is planning to slash the budget for its Reality Labs division by up to 30%, signaling a major shift away from its metaverse ambitions. The company is redirecting resources to accelerate its development of AI-powered glasses and other wearables.
Decoded:
The move follows staggering Reality Labs losses, which have surpassed $60 billion since 2020.
Meta is pivoting to tangible AI hardware and recently hired a former Apple design executive to lead a new creative studio for AI-powered wearables.
The budget cuts will likely lead to job losses and impact key metaverse projects, including the social VR platform Horizon Worlds.
Why It Matters: This decision marks a major retreat from the all-in metaverse strategy that defined Meta's 2021 rebrand. It highlights a broader industry pivot where companies are prioritizing near-term, tangible AI products over speculative virtual worlds.
Inside Anthropic's AI Workplace
Next in AI: Anthropic turned the lens inward, revealing in a detailed internal study that its engineers now use Claude for 60% of their work. The result is a self-reported 50% productivity boost and a new approach to tackling complex tasks.
Decoded:
Anthropic’s engineers report a 50% productivity increase from using Claude, a 2-3x jump from last year, with top use cases being debugging and understanding codebases.
Beyond just speed, AI is expanding the scope of work, with 27% of Claude-assisted tasks consisting of projects—like building new tools and conducting exploratory research—that otherwise would have been neglected.
The study also finds that engineers are becoming more “full-stack,” but it's not all upside, as some worry about losing deeper technical skills and having fewer opportunities for human collaboration while using tools like Claude Code.
Why It Matters: This study offers a rare, data-backed glimpse into how AI is already integrating into a high-skill workforce. It suggests the future of work isn't just about automation, but about reshaping roles and forcing us to reconsider career development and mentorship in an AI-augmented world.
NVIDIA's Job Forecast
Next in AI: NVIDIA CEO Jensen Huang pushes back on AI job destruction fears, using a famous failed prediction about radiologists to argue AI will augment professionals, not replace them, during a wide-ranging conversation about the future.
Decoded:
Huang pointed to Geoffrey Hinton's 2016 claim that AI would make radiologists obsolete within five years, a forecast that The New York Times recently revisited and analyzed.
Instead of disappearing, the number of radiologists grew because AI made image studying faster and more precise, allowing them to conduct more tests and improve hospital economics.
This supports Huang's popular view that you won't lose your job to an AI, but to someone who uses AI effectively.
Why It Matters: Huang's perspective shifts the conversation from replacement to augmentation, focusing on how AI automates tasks, not entire roles. This highlights the growing need for professionals in every field to adopt AI tools to stay competitive and increase their capabilities.
AI Pulse
OpenAI faces multiple lawsuits alleging that extended conversations with ChatGPT can induce "AI psychosis," a delusional disorder, with one suit claiming the bot reinforced a user’s misguided beliefs over 1,460 messages in 48 hours.
Geoffrey Hinton warned that "massive unemployment caused by AI" is very likely, arguing that tech giants are betting on replacing human workers to justify their trillion-dollar investments in data centers.
Samsung faces an internal supply chain rift, as soaring AI-driven demand for memory reportedly led its semiconductor division to reject a long-term DRAM deal from its own mobile division.
HHS unveiled its strategy to expand AI adoption across the department, focusing on improving efficiency, coordinating research, and promoting innovation in areas like drug development and patient data analysis.