The Download: a new Christian phone network, and debugging LLMs

A new US-wide cell phone network marketed to Christians is set to launch next week, featuring network-level controls that block porn and other content. The network, which can't be turned off even by adult account owners, also includes a filter on sexual content aimed at blocking material related to gender and trans issues, which is optional but turned on by default across all plans. However, the broad and subjective nature of these controls has raised concerns about the power of the network's founder to determine what content is allowed or banned.

The San Francisco–based startup Goodfire has released a new tool, Silico, that lets researchers peer inside an AI model and adjust its parameters during training. The goal is to make building AI models less like alchemy and more like a science. Using a technique called mechanistic interpretability, Silico maps the neurons and pathways inside a model and lets developers tweak them to reduce unwanted behaviors or steer outputs.

By exposing the 'knobs and dials,' Goodfire hopes to bring AI training closer to traditional software engineering. The National Science Foundation, a federal agency that funds major research projects to the tune of around $9 billion, has faced significant challenges in recent times. On Friday, the 22 scientists overseeing those efforts were all fired, marking another gut punch for science in the US.

Since 2025, the NSF has faced budget cuts, grant terminations, and mass firings, with staff numbers down sharply and many ambitious projects grinding to a halt. China's leading AI labs are taking a different approach to Silicon Valley AI companies, releasing 'open-weight' models that developers can download, adapt, and run on their own hardware. This approach went mainstream after DeepSeek open-sourced its R1 model, which matched top US systems at a fraction of the cost.

It also won something subtler: goodwill with developers. A growing cohort of Chinese labs is now following the same blueprint, making the future of AI more multipolar than Silicon Valley expected. In other news, Elon Musk has admitted that xAI trained Grok on OpenAI models, a practice known as 'distillation,' which is standard in AI but legally dubious.

A 'de-extinction' startup wants to resurrect a long-lost antelope using genomic editing, while an OpenAI model has outperformed ER doctors at diagnosing patients. Scientists are also exploring new ways to power AI data centers, including tiny nuclear reactors.