Quick Comparison: DeepSeek vs OpenAI (ChatGPT)
From X to Facebook, DeepSeek is everywhere. Nasdaq and BTC dumps blamed on it. So much FUD and hype. Let's fact-check what this 'magic' really is.
These past few days, DeepSeek has been flooding feeds from X to Facebook. Nasdaq dumps, BTC crashes — everyone's blaming it. So much FUD and storytelling to pump up the hype. For those of us who work with LLMs, DeepSeek isn't new. But for many others, it seems like magic. Let's fact-check what this magic really is:
1. Is DeepSeek smarter than OpenAI? ❌
No. In reality, DeepSeek is still playing catch-up with OpenAI. DeepSeek's r1 is roughly equivalent to OpenAI's o1, while OpenAI has already released o3 — which they claim approaches AGI, with near-human performance on the ARC-AGI benchmark.
So why do people think it's smarter? Simple: OpenAI charges money. How many people actually pay $20/month for o1 or $200/month for o1-pro? Most people use GPT-4o with limited reasoning capabilities. When they finally get to use r1 for free with advanced reasoning, of course it feels superior.
2. Is DeepSeek much cheaper than OpenAI? ✅
Absolutely. A month ago they released DeepSeek v3, matching GPT-4o at 1/11th the price — jaws dropped. Now with the r1 API matching o1 at 1/30th the price, you really can't fault the Chinese engineers' ability to deliver quality at low cost.
To twist the knife further, they're letting users access r1 on their web app for free. OpenAI got so spooked they announced they'll make o3-mini partially free (lol, closing the barn door after the horse has bolted).
3. Will DeepSeek destroy US tech? ✅❌
Yes and no.
Think of it this way: Nvidia's been selling excavators to gold miners, advertising them as limited edition. OpenAI used those excavators to dig up gold and sell it at premium prices. Then a neighbor comes along and digs up the same amount of gold... with just a few shovels, selling it at 1/30th the price. Those two should pack up shop, right?
But US tech going down entirely? Not quite. Give it 6 months. The big players who've been trailing in the AI race — Meta, Microsoft, Google — will copy DeepSeek's methodology. Everything is public and open-source, and the community is already trying to reproduce it. When that happens, whoever executes more efficiently wins — race to the bottom, users win.
4. Does DeepSeek change perceptions of China tech? ✅
Yes. If you were already impressed by TikTok's addictive recommendation algorithm, you might have had some appreciation. But DeepSeek r1 is on another level.
While US tech companies guard their models like secret sauce to squeeze money from users, DeepSeek came from behind and went big: fully open-source. Yes, 100% open-source.
Of course, the exact training data to replicate r1 isn't available (due to politically sensitive content about Taiwan, Tiananmen, etc.), but you can use any other SFT dataset.
Beyond open-source, they've also released several distilled models from the large r1 model to smaller Qwen and Llama models for local deployment... As someone who runs local LLMs, this gets a perfect 10 from me. 😍
5. Will DeepSeek and AI eliminate developers? ✅❌
Yes and no. This isn't a new topic — I've mentioned it several times before.
Friendly reminder for young people and parents: maybe consider something other than IT 🙂
o3 and r3 aren't far off. With reasoning capabilities through long chain of thought, they'll make any average engineer redundant (in the not-so-distant future).
But not all developers will be eliminated. Fundamentally, current models are trained through prompts to solve problems, so their problem-solving is excellent. But we'll still need developers who are good enough and understand business logic to frame the problems for AI to solve.
This is something AI can't replace (at least in the near future) because it lacks the holistic view, worldview, and business sense that humans have.
After developers, other professions will follow — especially jobs that are repetitive or require little creativity. Time to prepare.