DeepSeek V4 Drops, World Models Heat Up, and China Blocks Meta’s Manus Deal

Friday morning, I woke up to a flurry of notifications: DeepSeek had dropped a preview of V4, their next flagship model. And it’s a big deal for more reasons than one.

First off, the model can handle much longer prompts than the previous generation, thanks to a new architecture that processes large chunks of text more efficiently. That’s practical—anyone who’s tried to feed a novel-length document into an LLM knows the pain of context windows hitting their limit.

But here’s where it gets interesting: V4 matches the performance of leading closed-source models from Anthropic, OpenAI, and Google, while remaining open source. That alone would be noteworthy, but there’s a twist. This is DeepSeek’s first release optimized for Huawei’s Ascend chips—a direct test of China’s ability to sidestep Nvidia dependency. Given the ongoing US export restrictions, this is a strategic move as much as a technical one. I’m curious to see how the benchmarks hold up in real-world use, but the implications for the chip supply chain are hard to overstate.

Meanwhile, the race to build world models is heating up. The idea is simple: LLMs are great at text and code, but they stumble on physical tasks like folding laundry or navigating a busy street. World models aim to give AI a grounded understanding of how the physical world works—think physics, spatial reasoning, cause and effect. Stanford’s Fei-Fei Li and AMI Labs’ Yann LeCun are pushing this hard, arguing it’s the missing piece for robotics. I’ve seen this argument before, but the recent convergence of better sensors, cheaper compute, and advances in simulation makes it feel more tangible now. Whether it delivers on the promise of general-purpose robots remains to be seen, but it’s no longer just academic.

On the geopolitical front, China blocked Meta’s $2 billion acquisition of AI startup Manus, citing national security. Beijing called the deal a “conspiratorial” attempt to hollow out its tech base. This isn’t surprising—China has been tightening control over AI firms trying to exit, and the US-China rivalry is only intensifying. But I’d argue there are no winners in a tech cold war. The fragmentation hurts innovation everywhere.

Google also made waves, investing up to $40 billion in Anthropic at a $350 billion valuation. That’s a staggering number, reflecting the insatiable demand for compute and talent. Anthropic and OpenAI are essentially in an arms race for GPU capacity, and this deal gives Anthropic the firepower to keep up. I’m watching how this reshapes the competitive landscape—especially with Google’s own Gemini models in the mix.

And in a move that’s raising eyebrows across the scientific community, President Trump just fired the entire National Science Board. The NSF has been a cornerstone of US technology development, funding everything from early internet research to AI. The political interference worries me—it signals a willingness to weaponize science agencies, which could have long-term consequences for US leadership in R&D.

Finally, conspiracy theories about the Washington shooting are spreading like wildfire online, with over 300,000 posts already. It’s a grim reminder that misinformation thrives in moments of uncertainty, and platforms still haven’t solved the moderation problem.

World models are also featured on MIT Technology Review’s list of 10 Things That Matter in AI Right Now—worth checking out if you want a broader view of where the field is heading.