PLUS: Claude overtakes ChatGPT after its Pentagon dispute and why AI is making software engineering harder
Happy reading
Running a massive trillion-parameter model has typically been reserved for major data centers, but AMD is changing that. The company just showed how a small cluster of consumer PCs can run one of these powerful models locally.
This new method gives developers a path to work with cutting-edge AI without the high costs or data privacy concerns of the cloud. Does this breakthrough signal a new era of decentralized AI development, moving power away from the big tech monopolies?
In today’s Next in AI:
AMD’s guide for running 1T models on PCs
Claude overtakes ChatGPT after Pentagon dispute
Why AI is making software engineering harder
NVIDIA’s new AI for autonomous networks
The Trillion-Parameter Homebrew
Next in AI: In a major step for accessible AI, AMD has released a guide showing how to run a massive 1 trillion-parameter model using a small cluster of consumer-grade AI PCs.
Explained:
The setup links four AMD Ryzen AI Max+ PCs over a simple network, allowing them to function as a single, powerful AI accelerator.
A clever Linux kernel modification extends the usable GPU-addressable memory to 480GB across the cluster, enough to load the entire 375GB Kimi K2.5 model.
Performance optimizations like Flash Attention more than doubled text generation speeds on long prompts, increasing throughput from 3.46 tk/s to 8.30 tk/s.
Why It Matters:
This approach moves trillion-parameter AI from exclusive data centers to local, self-hosted environments. It gives developers and businesses a path to experiment with cutting-edge models while maintaining full data privacy and avoiding recurring cloud costs.
The Claude Rebellion
Next in AI: A public dispute with the Pentagon over AI safety has unexpectedly propelled its Claude app to #1 on the U.S. App Store, dethroning longtime rival ChatGPT.
Explained:
The conflict ignited after Anthropic refused to remove its safety guardrails that prevent Claude from being used for mass surveillance or autonomous lethal weapons, leading the Pentagon to blacklist the company.
In response, a user movement gained traction online, with many former ChatGPT users publicly canceling their subscriptions and documented their switch to Claude in a show of support for Anthropic's ethical stance.
Just hours after Anthropic was blacklisted, OpenAI announced its own Pentagon deal, which further fueled the user exodus from ChatGPT despite claims of similar safety limitations.
Why It Matters: This event demonstrates that a company's principles on AI safety can directly influence consumer choice and disrupt market leadership. The public's strong reaction signals that ethical transparency may become a key competitive advantage in the AI race.
The AI Productivity Paradox
Next in AI: AI coding assistants are accelerating code production, but a consensus is forming among engineers: these same tools are making the overall job of software engineering harder by increasing cognitive load and review burdens.
Explained:
AI helps experienced developers but can create a significant skill gap for new ones. One study found junior developers using AI assistants actually scored 17% lower on code comprehension tests than those who learned manually.
The productivity gains are not evenly distributed. Recent data shows a 3.6% quarterly productivity increase from AI tools, but the benefits went almost entirely to seniors with over six years of experience, while junior developers saw no net gain.
The role of a developer is fundamentally changing from a builder to a reviewer. Hiring reports show that developer roles now demand more system design and AI verification skills, as reviewing AI-generated code often takes more time and context than writing it from scratch.
Why It Matters:
The focus is shifting from code generation to code validation and high-level system design, placing a greater strain on engineering quality assurance. Companies must now adapt by investing in training for deep architectural thinking, not just faster coding, to build sustainable and skilled teams.
NVIDIA's Autopilot for Networks
Next in AI: NVIDIA is launching a new suite of tools to help telecom companies build autonomous networks. This collection of AI models and frameworks allows networks to manage and optimize themselves, paving the way for self-healing 5G and 6G infrastructure.
Explained:
The core is a new 30-billion-parameter Large Telco Model (LTM) based on NVIDIA's Nemotron 3 architecture. It's fine-tuned to understand telecom industry language, enabling it to reason through complex workflows like fault isolation and repair planning.
To help operators use the model, NVIDIA released an open source guide that teaches AI agents to think like network engineers. It uses examples of expert resolutions to train the AI on how to safely and effectively manage network operations.
New AI Blueprints put this intelligence into practice for specific tasks like reducing power consumption and optimizing network configurations. Early adopters like Telenor Group and Cassava Technologies are already using these tools to manage their complex network environments.
Why It Matters: This initiative pushes networks beyond simple automation, enabling them to intelligently anticipate and resolve issues without human intervention. Creating more resilient, efficient, and self-sufficient infrastructure is essential for handling the massive data demands of future AI applications and 6G.
AI Pulse
OpenAI raised $110 billion in new funding from Amazon, NVIDIA, and SoftBank, pushing its valuation to over $840 billion to scale its infrastructure and global product distribution.
Block announced plans to lay off 4,000 employees, with CEO Jack Dorsey explicitly citing AI as a key driver for restructuring the company into a leaner organization.
Researchers introduced a new spiking neural network language model that operates without an attention mechanism, achieving coherent multi-turn conversation on a low-power GPU.
The Plain Dealer began publishing articles drafted by an AI under the byline “Advance Local Express Desk” to cover local news, including public safety warnings and community events.