Friday brought a flurry of AI news, and I’m still sorting through it. Let me hit the highlights that actually matter.
DeepSeek V4: Three Reasons It’s Not Just Another Model
Chinese AI lab DeepSeek dropped a preview of V4, their long-awaited flagship, and it’s genuinely interesting. Not because it’s the best model ever—it’s not—but because of what it signals.
First, it handles much longer prompts than V3. They redesigned the architecture to chew through large text chunks more efficiently. That’s a practical win for anyone working with long documents or codebases.
Second, performance matches Anthropic’s Claude, OpenAI’s GPT-4, and Google’s Gemini on key benchmarks. For an open-source model, that’s impressive. Closed-source rivals still hold some edge, but the gap is narrowing faster than I expected.
Third—and this is the big one—V4 is DeepSeek’s first model optimized for Huawei’s Ascend chips. This is a direct test of China’s ability to sidestep US export controls on Nvidia hardware. If Ascend can handle cutting-edge training and inference at scale, Nvidia’s dominance gets a real challenge. I’ve been skeptical of Huawei’s chip ambitions, but this forces me to pay attention.
The model remains open-source, which keeps the pressure on everyone else. DeepSeek continues to punch above its weight.
The World Model Push: From Digital to Physical
Grace Huckins has a piece on world models, and it’s worth reading. The core argument: LLMs are great at text and code, but they suck at the physical world. Folding laundry, navigating a crowded street, picking up a fragile object—these remain hard.
Proponents like Fei-Fei Li and Yann LeCun argue that world models—systems that learn causal physics, spatial reasoning, and object permanence—are the missing piece. They’re not wrong. Current LLMs hallucinate because they lack grounded understanding of how the world works.
The race is on. Several labs are building world models for robotics. If they succeed, we’ll see robots that actually adapt to messy real-world environments instead of failing the moment something changes. That’s the promise. The reality is we’re still early, and the compute requirements are brutal.
China Blocks Meta’s Manus Acquisition
China’s regulators blocked Meta’s $2 billion buyout of AI startup Manus, citing national security. Beijing called the deal a “conspiratorial” attempt to hollow out its tech base. That’s strong language.
This isn’t surprising. China has been tightening control over AI firms trying to leave or sell to foreign buyers. The message is clear: your AI talent and IP stay here. The US-China AI rivalry just got another escalation point.
Google Drops $40 Billion on Anthropic
In a deal valuing Anthropic at $350 billion, Google is investing up to $40 billion. That’s not a typo. The funding will go toward computing infrastructure, which Anthropic desperately needs as it scales.
Anthropic and OpenAI are locked in a compute arms race. Both need massive clusters to train next-gen models. Google’s checkbook gives Anthropic a serious advantage—assuming regulators don’t block it. But with the current US administration, I wouldn’t bet on that.
Trump Fires the National Science Board
President Trump just fired the entire National Science Board. The NSF has been a backbone of US science funding, supporting everything from early internet research to AI. This move heightens fears of political interference in science.
The timing is awful. The US is already struggling to maintain its edge in AI research. Removing the board that oversees NSF grants doesn’t help. It sends a signal that science is political, which is dangerous for long-term innovation.
Quick Hits
- Conspiracy theories about the Washington shooting are spreading fast online. Platforms are struggling to keep up.
- China’s crackdown on AI firms trying to leave continues. TechCrunch has details.
- The US-China AI race is heating up, but as MIT Technology Review notes, there will be no winners in this competition.
That’s it for today. The AI landscape is shifting faster than ever. Keep your head up.
Comments (0)
Login Log in to comment.
Be the first to comment!