DeepSeek V4, World Models, and the Week AI Got Real

There’s a lot to unpack from the last few days in AI, and honestly, it’s the kind of week where you need to sit down with a coffee and just read.

First up, DeepSeek. On Friday, the Chinese firm released a preview of V4, their new flagship model. The headline feature is a new architecture that handles much longer prompts more efficiently—think processing a whole book instead of a chapter. That’s a practical win for anyone doing document-level work or long-form analysis.

What really caught my eye, though, is that V4 is the first DeepSeek model optimized for Huawei’s Ascend chips. This is a big deal. China has been scrambling to reduce its dependence on Nvidia, and this release is a real-world test of whether domestic hardware can compete. The model is still open source, and its performance matches Anthropic’s Claude, OpenAI’s GPT-4, and Google’s Gemini. If Ascend holds up, the chip landscape just got more interesting.

Meanwhile, the conversation around world models is heating up. The argument is simple: LLMs are great at text and code, but they’re terrible at physics. You can’t teach a robot to fold laundry by giving it a book on laundry. World models—systems that build an internal simulation of how the physical world works—are proposed as the missing piece. Fei-Fei Li and Yann LeCun are the loudest advocates, and they’ve got a point. I’ve seen enough robotics demos fail because the AI can’t handle a slightly crumpled towel.

On the geopolitical front, China blocked Meta’s $2 billion acquisition of AI startup Manus, citing national security. Beijing called it a “conspiratorial” attempt to hollow out its tech base. This is the latest escalation in the US-China AI rivalry, and it’s getting expensive. Google, meanwhile, is investing up to $40 billion in Anthropic, valuing the firm at $350 billion. That’s a lot of compute credits.

President Trump fired the entire National Science Board, which has historically funded foundational tech research. The move is raising alarms about political interference in US science. And conspiracy theories about a recent Washington shooting are spreading online, with over 300,000 posts on some platforms within hours.

It’s a lot. But the pattern is clear: AI is no longer just a tech story. It’s a geopolitical, economic, and cultural one. And we’re all just trying to keep up.