Deepseek: Hype or Hyperbole?

(you can easily filter previous issues by application!)

Hey friends - If you've been following tech news lately, you've probably heard about DeepSeek, the Chinese AI company that sent NVIDIA's stock (and my portfolio 😅) plummeting and has everyone talking.

But as with any hot tech topic, there's a lot of misinformation floating around. Let's bust some myths and get to what actually matters for you.

Myth #1: "DeepSeek built their model for just $5.6M"

Look, this number has been making headlines, but it's misleading. That $5.6M only covers the final training run. Think of it like the cost of the final performance of a Broadway show - it doesn't include months of rehearsals, theater rental, or actor salaries. The real investment is much higher.

Myth #2: "They must have broken the rules to do this"

Nope! They actually played by the rules. DeepSeek used legal H800 GPUs (the nerfed version of the H100s), not banned H100s. Their innovations were specifically designed around the H800's memory bandwidth limitations, suggesting compliance with export controls.

Myth #3: "DeepSeek has beaten OpenAI"

Not quite. While their R1 model can go toe-to-toe with OpenAI's o1, OpenAI has already shown off o3, which is more powerful. DeepSeek wins the efficiency race, not the capability one.

Myth #4: "They built everything from scratch"

Deepseek’s model likely learned from existing models like GPT-4 through a process called distillation. It's legal but might break some terms of service. Think of it like reverse-engineering a recipe from a famous restaurant.

Myth #5: "Using DeepSeek is automatically unsafe"

It's not black and white. Yes, the native DeepSeek app sends data to China, but you can use their model through American services like Perplexity, which keeps your data in the US.

Myth #6: "This kills NVIDIA's business"

Actually, cheaper AI could mean more people use it (like how cheaper phones led to more phone users), potentially increasing demand for NVIDIA's chips in the long run.

Myth #7: "This is terrible for US tech companies"

Plot twist: cheaper AI could actually help US tech companies by making it easier to build AI into their products. It's like how cheaper internet access helped tech companies grow.

What This (actually) Means For You

1. Free Alternative to Premium ChatGPT

If you're not paying for ChatGPT (meaning no access to their advanced reasoning models like o1 or o3), DeepSeek's R1 could be a powerful alternative. It's especially good at math, coding, and logical reasoning.

2. Privacy First

If you care about privacy concerns, avoid their native app (both web and mobile versions). Instead, access it through American services like Perplexity if you want to keep your data stateside.

3. Stick to What Works

Before jumping ship from your current AI tool, consider this: Unless DeepSeek can do something your current tool absolutely can't, the hassle of switching might not be worth it. Remember, productivity isn't just about having the newest tool - it's about having a smooth, reliable workflow.

The Bottom Line

DeepSeek is impressive and shows how AI is becoming more accessible. But don't feel pressured to switch just because it's the new hot thing. Evaluate it based on your specific needs, privacy concerns, and existing workflow.

When you’re ready, here are all the ways how I can help.

Was this forwarded to you? Join Workspace Essentials for free 😉