Meet Alpha: Your personal AI Investment Assistant

Built by the investing platform Public, Alpha keeps a sharp eye on the stocks, ETFs, and crypto that matter most to you.

Here’s why you’ll love it:

• Stay in the know: Real-time alerts on the market activity that matters to you.

• Contextual insights: Alpha doesn’t just tell you when something changes—it tells you why.

• Natural-language interface: Ask Alpha anything and get clear answers on trends, earnings, and historical data.

Ready to take control of your investments? Download Alpha on iOS and start your investment watchlist for just $1/week—or enjoy it for free if you're a Public member.

Disclaimer: Alpha is an AI research tool powered by GPT-4.  Alpha is experimental and may generate inaccurate responses.  Output from Alpha should not be construed as investment research or recommendations, and should not serve as the basis for any investment decision. Public makes no warranties about its accuracy, completeness, quality, or timeliness of any Alpha out. Please independently evaluate and verify any such output for your own use case.

China’s DeepSeek V3 takes on AI’s biggest names

Key Points 

• DeepSeek V3 outperforms major AI models like Meta’s Llama 3.1 and OpenAI’s GPT-4o in benchmarks.

• Built with 671 billion parameters, it’s 1.6 times larger than Meta’s Llama 3.1.

👨🏻‍💻News - A Chinese lab has unveiled one of the most powerful open AI models yet: DeepSeek V3. Developed by DeepSeek, the model launched last week under a permissive license that lets developers download, modify, and even use it for commercial purposes.

🕵🏻‍♂️How does it compare? DeepSeek V3 excels in tasks like coding, translating, and writing essays from descriptive prompts. Internal benchmarks place it ahead of both open and closed models, including Meta’s Llama 3.1 405B, OpenAI’s GPT-4o, and Alibaba’s Qwen 2.5 72B. 

Notably, it dominated Codeforces programming competitions and Aider Polyglot, a test for writing new code that integrates seamlessly with existing projects.

🤓What’s more? DeepSeek V3 is built on a massive 14.8-trillion-token dataset, representing nearly 11 trillion words. Its architecture includes 671 billion parameters (685 billion on Hugging Face), making it 1.6 times larger than Meta’s Llama 3.1. While bigger models often perform better, they also require more powerful hardware. Running DeepSeek V3 at its best needs a setup of high-end GPUs.

Even so, DeepSeek managed to train this model in just two months using Nvidia H800 GPUs—despite U.S. export restrictions—and spent only $5.5 million, a fraction of what’s typically required for models of this scale.

ByteDance skirts U.S. restrictions to buy $7 billion in Nvidia chips in 2025

Key Points 

• The company sidesteps restrictions by storing chips in data centers outside the U.S., like in Southeast Asia.

☕News - TikTok parent company ByteDance is planning a major purchase of Nvidia chips in 2025, despite ongoing U.S. restrictions on Chinese companies acquiring American AI hardware. 

ByteDance intends to spend $7 billion on Nvidia’s high-performance chips, potentially making the company one of the largest owners of these sought-after components. 

The move is seen as an effort to strengthen ByteDance's AI capabilities as competition in the global tech market heats up. Despite U.S. efforts to block Chinese access to such technology, ByteDance appears undeterred in its ambitions.

➿The loophole that changes everything - ByteDance is circumventing the restrictions by using a strategic loophole. Instead of directly importing the chips to China, the company plans to store them in data centers located in regions outside the U.S., such as Southeast Asia. 

ByteDance has stated that it is fully compliant with U.S. export control rules, adding, “ByteDance has not bought H100s for its data centers outside of the U.S. since the relevant U.S. export control rules took effect.” This approach technically adheres to U.S. regulations, as the chips are not being used within China.